CCNA Describe Fundamental Principles Of Machine Learning On Azure Questions — Page 2 of 3

MCQhard

A bank uses a machine learning model to predict credit card fraud. The model's output is a probability score. The business wants to minimize the number of false positives (legitimate transactions incorrectly flagged as fraud) because these cause customer dissatisfaction. At the same time, they must also catch most fraudulent transactions. Which metric should the bank optimize to balance these two goals?

A.A: Accuracy

B.B: Precision

C.C: Recall

D.D: F1 score

AnswerD

F1 score is the harmonic mean of precision and recall, so it provides a single balanced measure of both completeness and trustworthiness of fraud predictions. In credit card fraud modeling, both undetected fraud (false negatives) and blocked legitimate purchases (false positives) are costly, and F1 score specifically rewards models that achieve a strong trade-off between the two. It is the correct metric for this scenario because it avoids the misleading optimism of accuracy on imbalanced data.

Why this answer

The F1 score is the harmonic mean of precision and recall, making it the ideal metric when a balance between minimizing false positives (precision) and catching most fraudulent transactions (recall) is required. In this credit card fraud detection scenario, optimizing F1 ensures the model reduces customer dissatisfaction from false positives while still maintaining high detection of actual fraud.

Exam trap

The trap here is that candidates often choose precision or recall alone, not realizing that the F1 score is specifically designed to balance both metrics when the business requires minimizing false positives while still catching most true positives.

How to eliminate wrong answers

Option A is wrong because accuracy measures overall correct predictions (true positives + true negatives divided by total predictions) and can be misleading in imbalanced datasets like fraud detection, where legitimate transactions vastly outnumber fraudulent ones; a model that always predicts 'not fraud' could achieve high accuracy but fail to catch any fraud. Option B is wrong because precision focuses solely on the proportion of flagged transactions that are actually fraudulent (true positives / (true positives + false positives)), which minimizes false positives but does not account for missed fraudulent transactions (false negatives), potentially allowing many frauds to go undetected. Option C is wrong because recall (sensitivity) measures the proportion of actual fraudulent transactions correctly identified (true positives / (true positives + false negatives)), which prioritizes catching fraud but can lead to a high number of false positives, directly conflicting with the business goal of minimizing customer dissatisfaction.

Practice this question →

MCQmedium

A data scientist has a dataset with 100 features and 10,000 samples. They want to reduce the number of features while retaining as much variance as possible, to improve model training speed and reduce overfitting. Which technique should they use?

A.Feature scaling

B.Principal Component Analysis (PCA)

C.Regularization

D.Cross-validation

AnswerB

Principal Component Analysis (PCA) is an unsupervised linear dimensionality reduction technique that computes the eigenvectors of the covariance matrix and sorts them by their eigenvalues, representing the amount of explained variance. Projecting the data onto the top k principal components yields k orthogonal composite features that retain the most variance, directly reducing the feature count from 100 to k. In Azure Machine Learning, the PCA module performs this projection as a preprocessing step, making it the appropriate choice for this scenario.

Why this answer

Principal Component Analysis (PCA) is an unsupervised dimensionality reduction technique that transforms the original features into a new set of orthogonal components, ordered by the amount of variance they capture. By selecting only the top principal components, the data scientist can significantly reduce the feature count (e.g., from 100 to 20) while retaining the majority of the dataset's variance, which directly improves model training speed and reduces overfitting.

Exam trap

The trap here is that candidates often confuse regularization (which reduces overfitting by shrinking coefficients) with dimensionality reduction, or they think feature scaling alone can reduce feature count, when PCA is the correct technique for explicitly reducing the number of features while preserving variance.

How to eliminate wrong answers

Option A is wrong because feature scaling (e.g., standardization or normalization) adjusts the range of feature values but does not reduce the number of features; it is often a preprocessing step before applying PCA, not a dimensionality reduction technique itself. Option C is wrong because regularization (e.g., L1 or L2) penalizes model coefficients to prevent overfitting but does not reduce the number of features in the dataset; it works during model training, not as a preprocessing step. Option D is wrong because cross-validation is a model evaluation technique used to assess generalization performance by splitting data into training and validation folds; it does not reduce feature count or variance retention.

Practice this question →

MCQmedium

What is 'ONNX' and why is it relevant to Azure AI?

A.An Azure-specific machine learning programming language

B.An open model interchange format enabling models to move between frameworks and edge deployments

C.A database for storing machine learning model training data

D.A Microsoft cloud service for distributed model training

AnswerB

ONNX (Open Neural Network Exchange) is an open interchange format for representing a trained ML model as a computational graph with standardized operators, learned parameters, and metadata. This allows a model developed in one framework (e.g., PyTorch) to be converted and run in another runtime (e.g., ONNX Runtime, TensorRT, or Windows ML) with minimal changes, and to be deployed consistently from cloud clusters down to resource-constrained edge devices. Because the format itself is framework- and cloud-agnostic, teams can train once and make the model portable across numerous inference engines, avoiding lock-in to a single vendor. That matches the description in the question exactly.

Why this answer

ONNX (Open Neural Network Exchange) is an open-source model interchange format that allows machine learning models to be transferred between different frameworks (e.g., PyTorch, TensorFlow, scikit-learn) and deployed across various environments, including edge devices. In Azure AI, ONNX is relevant because it enables interoperability and portability, allowing models trained in one framework to be optimized and run efficiently using Azure's ONNX Runtime, which accelerates inference on both cloud and edge hardware.

Exam trap

The trap here is that candidates confuse ONNX with a proprietary Azure service or a programming language, when in fact it is an open, cross-platform model interchange format designed for portability and not tied to any single cloud provider.

How to eliminate wrong answers

Option A is wrong because ONNX is not a programming language; it is a serialized model format, and Azure-specific ML languages include languages like Python or R, not ONNX. Option C is wrong because ONNX does not store training data; it stores model architecture and weights, while databases like Azure SQL or Cosmos DB are used for data storage. Option D is wrong because ONNX is not a cloud service for distributed training; Azure offers services like Azure Machine Learning for distributed training, while ONNX is purely an interchange format.

Practice this question →

MCQmedium

A data scientist trains a regression model to predict house prices using features like bedrooms, square footage, and location. The model achieves a low error on the training data but performs significantly worse when used to predict prices in a new city with different property characteristics. Which concept best explains this poor performance?

A.Underfitting

B.Overfitting

C.Data leakage

D.Bias-variance tradeoff

AnswerB

Overfitting occurs when a model with excessive capacity learns not only the genuine patterns in the training data but also its random noise and idiosyncrasies. This results in very low training error but poor generalization, and the effect becomes especially visible when new houses come from a slightly different distribution, because the learned noise no longer matches. The symptom described—good performance during training but degraded performance on new data—is the classic signature of overfitting.

Why this answer

The model performs well on training data but poorly on new data from a different city, which is the classic symptom of overfitting. Overfitting occurs when a model learns noise and specific patterns in the training data that do not generalize to unseen data, especially when the new data has different characteristics (e.g., different property market dynamics). In this case, the model has memorized the training city's price patterns rather than learning generalizable relationships.

Exam trap

The trap here is that candidates may confuse overfitting with the bias-variance tradeoff, but the question specifically asks for the concept that best explains the poor performance on new data, which is overfitting, not the general tradeoff.

How to eliminate wrong answers

Option A is wrong because underfitting would result in high error on both training and test data, not low training error and high test error. Option C is wrong because data leakage involves using information from the test set during training, which would artificially inflate training performance but is not described here; the issue is generalization to a new city, not a data contamination problem. Option D is wrong because while bias-variance tradeoff is a related concept, it does not specifically name the phenomenon; overfitting is the direct explanation for low training error and high test error on new data.

Practice this question →

MCQmedium

What is 'semi-supervised learning' and when is it useful?

A.Training a model that is partially supervised by one human and partially by another

B.Using small amounts of labelled data alongside large amounts of unlabelled data to train a model

C.A model that receives feedback from users during deployment to improve over time

D.Training that automatically stops halfway through and resumes the next day

AnswerB

This is the core definition of semi-supervised learning. The model first learns patterns from the abundant unlabeled examples, then uses the scarce labeled examples to anchor those patterns to the correct output classes, often through self-training or pseudo-labelling. It is especially valuable when labelling is expensive, because unlabeled data is usually plentiful and cheap, letting the model improve decision boundaries without fully labeled datasets.

Why this answer

Semi-supervised learning combines a small set of labeled data with a large set of unlabeled data to train a model. This approach is useful when labeling data is expensive or time-consuming, but large volumes of unlabeled data are readily available. The model first learns patterns from the labeled subset, then propagates those labels to the unlabeled data, iteratively improving its accuracy.

Exam trap

The trap here is that candidates confuse semi-supervised learning with active learning or human-in-the-loop workflows, but the key differentiator is the use of both labeled and unlabeled data in the training process, not the number of humans or feedback loops.

How to eliminate wrong answers

Option A is wrong because it describes a human workflow (multiple labelers), not a machine learning paradigm; semi-supervised learning refers to the data labeling strategy, not the number of human supervisors. Option C is wrong because it describes online learning or reinforcement learning, where the model updates from live user feedback, not the semi-supervised combination of labeled and unlabeled data. Option D is wrong because it describes checkpointing or resumable training, which is a fault-tolerance mechanism, not a learning paradigm.

Practice this question →

MCQmedium

What is 'overfitting' in machine learning and how does Azure ML help prevent it?

A.When a model is trained on too much data and becomes too accurate

B.When a model learns training data too specifically and fails to generalise to new data

C.When a model's predictions exceed the acceptable numerical range

D.When Azure ML runs training for longer than the allocated compute budget

AnswerB

Overfitting is defined by the gap between training and test performance: the model acquires such a detailed mapping of training examples, including outliers and noise, that it scores near perfectly on those examples but loses the ability to generalise to unseen inputs. It effectively memorises the training set instead of learning the broader patterns that would let it make reliable predictions on new data.

Why this answer

Overfitting occurs when a machine learning model learns the training data too precisely, including noise and outliers, resulting in poor performance on unseen data. Azure ML helps prevent overfitting through automated machine learning (AutoML) which applies regularization, cross-validation, and early stopping techniques, as well as by enabling easy configuration of train/test splits and hyperparameter tuning.

Exam trap

The trap here is that candidates confuse overfitting with high accuracy or large datasets, but the key is that overfitting is about poor generalization, not just high performance on training data.

How to eliminate wrong answers

Option A is wrong because overfitting is not caused by training on too much data; in fact, more data often reduces overfitting. Option B is correct as described. Option C is wrong because exceeding an acceptable numerical range describes prediction errors or data normalization issues, not overfitting.

Option D is wrong because exceeding a compute budget is a resource constraint, not a machine learning concept related to model generalization.

Practice this question →

MCQmedium

What is a training job in Azure Machine Learning?

A.A batch prediction job that scores new data against a deployed model

B.A single execution of a training script that produces a trained model and tracked metrics

C.A scheduled report on model performance in production

D.A data preprocessing pipeline that cleans raw datasets

AnswerB

A training job is the core unit of model training in Azure ML: it is a single, tracked execution of a training script on a chosen compute target, and that script performs the actual fitting of a model to data. During the run, Azure ML captures parameters, logs, and metrics (for example via MLflow or run.log_metric), and when complete it produces a model artifact that can be registered or evaluated. This aligns exactly with the definition of a training job, distinguishing it from monitoring, preprocessing, and batch-scoring workloads.

Why this answer

A training job in Azure Machine Learning is a single execution of a training script that runs on a specified compute target, producing a trained model and logging metrics, parameters, and artifacts. This is the fundamental unit of model training in Azure ML, distinct from batch inference or data preprocessing.

Exam trap

The trap here is confusing the training job with other Azure ML workflow steps like batch inference, monitoring, or data preprocessing, which are separate job types with distinct purposes and outputs.

How to eliminate wrong answers

Option A is wrong because a batch prediction job that scores new data against a deployed model is an inference or scoring job, not a training job. Option C is wrong because a scheduled report on model performance in production is a monitoring or evaluation task, not a training job. Option D is wrong because a data preprocessing pipeline that cleans raw datasets is a data preparation step, which may precede training but is not itself a training job.

Practice this question →

MCQeasy

What is 'model deployment' in Azure Machine Learning?

A.Uploading training data to Azure Blob Storage for model training

B.Making a trained model available as a callable endpoint for applications to use

C.Distributing the training job across multiple compute nodes

D.Publishing a model to the Azure Marketplace for other organisations to purchase

AnswerB

Deployment is the process of taking a trained and registered model and hosting it as an inferencing service—typically a REST endpoint—that applications can invoke with JSON payloads to receive predictions. Azure Machine Learning managed online endpoints handle model versioning, traffic splitting, authentication, and autoscaling for real-time scenarios; batch endpoints serve asynchronous, large-scale scoring. This is exactly what makes a model accessible for real-world use after the training phase is complete.

Why this answer

Model deployment in Azure Machine Learning is the process of taking a trained model and hosting it as a web service endpoint (e.g., via Azure Kubernetes Service or Azure Container Instances) so that applications can send data to it and receive predictions in real time or batch mode. This makes the model operational and accessible for inference, which is the core purpose of deployment.

Exam trap

The trap here is that candidates confuse 'model deployment' with other stages of the ML lifecycle, such as data preparation (Option A) or training optimization (Option C), because they focus on the word 'model' rather than the specific action of making it available for inference.

How to eliminate wrong answers

Option A is wrong because uploading training data to Azure Blob Storage is a data ingestion step, not model deployment; deployment involves hosting the trained model, not storing raw data. Option C is wrong because distributing training across multiple compute nodes is a parallel training or distributed computing technique, not deployment; deployment focuses on serving the model after training. Option D is wrong because publishing a model to the Azure Marketplace is a commercial distribution action, not a technical deployment; Azure Machine Learning deployment creates a callable endpoint, not a marketplace listing.

Practice this question →

MCQmedium

What is a neural network?

A.A computer network for distributed AI training across multiple servers

B.A machine learning model architecture with layers of interconnected nodes that learn representations

C.A database for storing trained ML models

D.A rule-based expert system for decision making

AnswerB

A neural network is a machine learning architecture composed of layers of interconnected nodes, or neurons, where each connection has an adjustable weight. Activations flow through the layers, and with non-linear activation functions the network can learn hierarchical representations from data. Training methods such as backpropagation and gradient descent update these weights so the model can identify complex patterns that would be difficult to encode explicitly.

Why this answer

A neural network is a machine learning model architecture composed of layers of interconnected nodes (neurons) that process input data through weighted connections and activation functions. These layers learn hierarchical representations of data, enabling the model to capture complex patterns and relationships without explicit rule-based programming. This aligns with option B as the correct definition.

Exam trap

The trap here is that candidates confuse the term 'network' in 'neural network' with a computer network or distributed system, leading them to incorrectly select option A.

How to eliminate wrong answers

Option A is wrong because a neural network is not a computer network for distributed AI training; distributed training across multiple servers is a technique (e.g., using Azure Machine Learning with Horovod or PyTorch DistributedDataParallel), not the definition of a neural network itself. Option C is wrong because a neural network is a model architecture, not a database; storing trained ML models is done in model registries like Azure Machine Learning model registry or container registries. Option D is wrong because a neural network learns from data via backpropagation and gradient descent, unlike a rule-based expert system that relies on hardcoded if-then rules and does not learn representations.

Practice this question →

MCQmedium

What is 'ROC-AUC' and when is it a better metric than accuracy for classification?

A.ROC-AUC is always better than accuracy regardless of the use case

B.A threshold-agnostic metric that measures discrimination ability — better than accuracy for imbalanced classes

C.A metric specifically for measuring multi-class classification across more than two classes

D.An evaluation metric only applicable to models trained on Azure Machine Learning

AnswerB

ROC-AUC evaluates how well a model separates positive and negative classes by measuring the probability that a randomly chosen positive instance is ranked higher than a randomly chosen negative instance, across every possible decision threshold. Because it aggregates the true positive rate and false positive rate over all thresholds, it does not require picking a single cutoff and is therefore threshold-agnostic. In imbalanced datasets, accuracy can be misleadingly high by simply predicting the majority class, whereas ROC-AUC reflects the model's ranking ability and stays informative even when one class dominates.

Why this answer

ROC-AUC (Receiver Operating Characteristic - Area Under the Curve) is a threshold-agnostic metric that measures a model's ability to discriminate between positive and negative classes across all possible classification thresholds. It is a better metric than accuracy when dealing with imbalanced classes because accuracy can be misleadingly high if the model simply predicts the majority class, whereas ROC-AUC evaluates the trade-off between true positive rate and false positive rate independently of class distribution.

Exam trap

The trap here is that candidates often assume accuracy is always the best metric, failing to recognize that ROC-AUC is specifically designed to evaluate model performance independently of class imbalance, which is a common scenario tested in AI-900.

How to eliminate wrong answers

Option A is wrong because ROC-AUC is not always better than accuracy; for balanced datasets with equal misclassification costs, accuracy is often simpler and more interpretable. Option C is wrong because ROC-AUC is fundamentally a binary classification metric; while extensions like macro-averaged or micro-averaged ROC-AUC exist for multi-class problems, the standard definition applies to two classes only. Option D is wrong because ROC-AUC is a general machine learning evaluation metric that can be computed for any binary classifier, regardless of the platform (Azure, AWS, on-premises, etc.).

Practice this question →

MCQmedium

A hospital deploys a machine learning model to screen patients for a rare disease. Only 0.1% of patients actually have the disease. The model correctly identifies most positive cases but also flags many healthy patients as potentially having the disease. The hospital wants to minimize the number of healthy patients who are incorrectly told they might have the disease. Which metric should the model optimize?

A.Recall

B.Precision

C.F1 score

D.Accuracy

AnswerB

Precision (TP/(TP+FP)) measures the fraction of positive predictions that are actually correct. Maximizing precision directly reduces false positives, meaning that when the model flags a patient for screening, it is likely a real case, sparing healthy patients from needless follow-up tests and worry. Given the hospital's explicit goal to minimize false positives, precision is the metric that aligns with that clinical priority.

Why this answer

Precision measures the proportion of positive identifications that are actually correct. In this scenario, the hospital wants to minimize false positives (healthy patients incorrectly told they might have the disease). Optimizing precision directly reduces false positives, which is the stated goal.

Exam trap

The trap here is that candidates often default to 'Accuracy' for imbalanced datasets or 'Recall' for medical screening, but the question explicitly asks to minimize false positives, which directly points to Precision as the correct metric.

How to eliminate wrong answers

Option A (Recall) is wrong because recall measures the proportion of actual positives correctly identified; optimizing recall would reduce false negatives (missing diseased patients), but the hospital's priority is minimizing false positives, not false negatives. Option C (F1 score) is wrong because F1 is the harmonic mean of precision and recall; while it balances both, it does not specifically minimize false positives—it trades off between precision and recall, which may still allow many false positives if recall is prioritized. Option D (Accuracy) is wrong because accuracy measures overall correct predictions; with a highly imbalanced dataset (0.1% disease prevalence), a model that always predicts 'no disease' would achieve 99.9% accuracy but would fail to identify any positive cases and would not address the false positive minimization goal.

Practice this question →

MCQeasy

A data scientist is preparing a dataset to train a model that predicts customer churn. The dataset includes a column 'CustomerID' which is a unique identifier for each customer. Should the data scientist include the 'CustomerID' column as a feature in the training data?

A.Yes, because it uniquely identifies each customer and helps the model differentiate them.

B.No, because the CustomerID is a random unique identifier with no predictive power for churn.

C.Yes, because the model can learn patterns from the numeric values.

D.No, because the CustomerID column contains too many missing values.

AnswerB

The CustomerID is an arbitrary, randomly assigned string that carries no information about a customer's behavior, demographics, or service usage. A churn model learns by identifying correlations between input features and the target variable; because CustomerID has no systematic relationship with churn, the algorithm would simply memorize the ID-to-outcome mapping in the training data. This memorization leads to overfitting and poor performance on new, unseen customers. Therefore, the correct answer is No.

Why this answer

CustomerID is a unique identifier that does not contain any meaningful pattern or relationship with the target variable (churn). Including such a column would introduce noise and risk overfitting, as the model could memorize each ID rather than learning generalizable patterns. In Azure Machine Learning, features should be predictive attributes, not arbitrary labels.

Exam trap

The trap here is that candidates may think unique identifiers are useful for differentiation, but the exam tests the principle that features must have predictive power and that arbitrary IDs introduce noise rather than signal.

Why the other options are wrong

CustomerID is a unique identifier with no correlation to churn; including it would cause overfitting as the model would memorize IDs rather than learn generalizable patterns.

CustomerID is a unique identifier with no inherent relationship to churn; including it would cause overfitting as the model would memorize IDs rather than learn generalizable patterns. Numeric values of IDs are arbitrary and carry no predictive power.

The question states the dataset includes a 'CustomerID' column, but does not mention missing values. The reason to exclude it is its lack of predictive power, not missing data.

Practice this question →

MCQmedium

A data scientist trains a machine learning model to predict house prices based on features like square footage, number of bedrooms, and location. The model achieves a very low error on the training data but performs poorly on a held-out test set. Which term best describes this situation?

A.Underfitting

B.Overfitting

C.High bias

D.High variance

AnswerB

Overfitting means the model has effectively memorized the training examples, including their random noise and idiosyncrasies, rather than learning a generalizable pattern. As a result, it achieves near-perfect training accuracy but performs poorly on unseen test data because the test set does not contain those same noise patterns. This direct training-versus-test performance gap is the classic signature of overfitting.

Why this answer

The model performs exceptionally well on training data but poorly on test data, which is the classic symptom of overfitting. Overfitting occurs when the model learns noise and specific patterns in the training set rather than generalizing to unseen data. In Azure Machine Learning, this can be detected by monitoring the gap between training and validation metrics, and mitigated using techniques like regularization or early stopping.

Exam trap

The trap here is that candidates confuse 'high variance' (the cause) with 'overfitting' (the observed behavior), but the question asks for the term that best describes the situation, not the underlying statistical property.

How to eliminate wrong answers

Option A is wrong because underfitting describes a model that performs poorly on both training and test data due to insufficient learning capacity, not the high training accuracy seen here. Option C is wrong because high bias typically leads to underfitting, where the model oversimplifies and misses important patterns, resulting in high error on both sets. Option D is wrong because high variance is a cause of overfitting, but the term 'overfitting' itself is the correct descriptor for the situation where the model fits training data too closely and fails on test data.

Practice this question →

MCQmedium

Which metric is MOST appropriate for evaluating a regression model's performance?

A.Accuracy

B.Root Mean Squared Error (RMSE)

C.Precision and recall

D.AUC-ROC curve

AnswerB

Root Mean Squared Error (RMSE) is the standard evaluation metric for regression models. It is calculated by taking the square root of the average of the squared differences between predicted and actual values, which penalizes large errors more heavily than small ones. A lower RMSE indicates predictions are closer to the true continuous values, and it is expressed in the same units as the target variable, making it directly interpretable. Unlike classification metrics, RMSE naturally handles continuous numeric predictions and is sensitive to outliers, which can be desirable when large errors are especially problematic.

Why this answer

Root Mean Squared Error (RMSE) is the most appropriate metric for evaluating a regression model because it measures the average magnitude of prediction errors in the same units as the target variable, penalizing larger errors more heavily due to squaring. In Azure Machine Learning, regression models like Linear Regression or Decision Forest Regression are evaluated using RMSE to quantify how well the predicted continuous values match actual values.

Exam trap

The trap here is that candidates often confuse regression and classification metrics, mistakenly applying Accuracy (a classification metric) to regression problems because they think it measures 'correctness' in a general sense, without understanding that regression requires error-based metrics like RMSE.

How to eliminate wrong answers

Option A is wrong because Accuracy is a classification metric that measures the proportion of correct predictions out of total predictions, and it is not suitable for regression tasks where the output is a continuous value rather than a discrete class. Option C is wrong because Precision and recall are classification metrics used to evaluate the performance of binary or multiclass classifiers, focusing on true positives and false positives/negatives, not continuous predictions. Option D is wrong because AUC-ROC curve is a classification metric that plots the true positive rate against the false positive rate at various threshold settings, and it does not apply to regression models which predict continuous outcomes.

Practice this question →

MCQmedium

A data scientist is building a binary classification model to predict fraudulent credit card transactions. The dataset is highly imbalanced: only 1% of transactions are fraudulent. The cost of a false negative is very high because missing a fraudulent transaction can lead to significant financial loss. Which evaluation metric should the data scientist prioritize to minimize false negatives?

A.Accuracy

B.Precision

C.Recall

D.F1 Score

AnswerC

Recall is defined as true positives divided by the sum of true positives and false negatives (TP / (TP + FN)). It directly measures the proportion of actual positive cases the model successfully captures, so maximizing recall is the most straightforward metric when the business goal is to minimize false negatives. In this scenario, missing a positive case is more costly than flagging a false positive, making recall the correct choice.

Why this answer

Recall (also known as sensitivity or true positive rate) measures the proportion of actual positive cases (fraudulent transactions) that are correctly identified. In this highly imbalanced scenario where missing a fraud (false negative) is extremely costly, maximizing recall ensures that the model catches as many fraudulent transactions as possible, even if it means some false positives occur. This directly aligns with the goal of minimizing false negatives.

Exam trap

The trap here is that candidates often choose Accuracy because it is the most intuitive metric, failing to recognize that in imbalanced datasets with high false-negative cost, recall is the critical measure to minimize missed positives.

Why the other options are wrong

Accuracy is misleading in imbalanced datasets because a model that predicts all transactions as legitimate would achieve 99% accuracy but fail to detect any fraud, which does not minimize false negatives.

Precision focuses on minimizing false positives, not false negatives. In this scenario, the high cost of false negatives means recall is the priority.

F1 Score balances precision and recall, but in this scenario where minimizing false negatives is critical, recall is the direct metric to optimize. F1 Score would penalize a model that achieves high recall at the expense of precision, which is acceptable here.

Practice this question →

MCQmedium

A data scientist wants to train a machine learning model to predict the exact market price of a house based on features such as square footage, number of bedrooms, and location. Which type of machine learning task should be used?

A.Classification

B.Regression

C.Clustering

D.Anomaly Detection

AnswerB

Regression predicts a continuous numeric value, so it fits forecasting an exact house price from features such as square footage, bedrooms and location. Classification would output discrete labels or categories, which cannot represent the precise market price the scenario requires.

Why this answer

Predicting the exact market price of a house is a regression task because the target variable (price) is a continuous numeric value. Regression algorithms, such as linear regression or decision tree regression, learn the relationship between input features (e.g., square footage, bedrooms, location) and a continuous output. In Azure Machine Learning, you would select a regression model from the designer or AutoML to solve this problem.

Exam trap

The trap here is that candidates confuse regression with classification because both involve supervised learning, but regression outputs a continuous number while classification outputs a discrete label.

Why the other options are wrong

The question asks for predicting an exact market price, which is a continuous numeric value. Classification predicts discrete categories or labels, not continuous values.

Clustering is an unsupervised learning task used to group similar data points, but this question requires predicting a continuous numeric value (market price), which is a supervised regression problem.

Anomaly detection identifies rare or unusual data points, not continuous values like house prices. The goal here is to predict a specific numeric price, which is a regression task.

Practice this question →

MCQmedium

A data scientist trains a binary classification model to detect fraudulent credit card transactions. The dataset contains 99.5% legitimate transactions and 0.5% fraudulent transactions. The model predicts every transaction as legitimate and achieves 99.5% accuracy on the test set. Which metric would best reveal that the model is failing to identify any fraudulent transactions?

A.Precision

B.Recall

C.F1 score

D.Mean Absolute Error (MAE)

AnswerB

Recall, or sensitivity, is the ratio of true positives to all actual positives (true positives + false negatives). With the model predicting no fraud, true positives are 0 while false negatives equal the total number of actual fraudulent transactions, yielding a recall of 0. This directly measures the model's inability to catch any positive cases, which is exactly the failure mode described in the scenario, so recall is the most appropriate metric to highlight the problem.

Why this answer

Recall (also known as sensitivity) measures the proportion of actual positive cases correctly identified by the model. In this scenario, the model predicts all transactions as legitimate, so it correctly identifies zero fraudulent transactions, giving a recall of 0%. Accuracy alone is misleading because the dataset is highly imbalanced (99.5% legitimate, 0.5% fraudulent), and a 99.5% accuracy can be achieved by simply predicting the majority class.

Recall directly reveals the model's failure to detect any fraud.

Exam trap

The trap here is that candidates see 99.5% accuracy and assume the model is performing well, failing to recognize that accuracy is a poor metric for imbalanced datasets and that recall specifically measures the model's ability to find positive cases (fraud).

How to eliminate wrong answers

Option A is wrong because precision measures the proportion of predicted positive cases that are actually positive; if the model predicts no positives, precision is undefined (division by zero) or 0/0, but it does not directly show the failure to find actual fraud. Option C is wrong because the F1 score is the harmonic mean of precision and recall; if recall is 0, the F1 score is 0, but the F1 score is a combined metric and does not isolate the failure to detect fraud as directly as recall does. Option D is wrong because Mean Absolute Error (MAE) is a regression metric that measures average absolute error between predicted and actual continuous values; it is not applicable to binary classification tasks like fraud detection.

Practice this question →

MCQmedium

A data scientist is developing a classification model to detect fraudulent transactions. The dataset is split into training and test sets. The data scientist repeatedly tunes the model's hyperparameters and evaluates performance on the test set until the test accuracy reaches 95%. However, when the model is deployed on new, unseen data, its accuracy drops to 70%. Which concept best explains this performance degradation?

A.Overfitting to the training data

B.Data leakage from the training set to the test set

C.Overfitting to the test set

D.Underfitting the training data

AnswerC

This is correct. When hyperparameters are tuned by repeatedly evaluating against the same test set, the model selection process implicitly 'fits' the test set just as a gradient descent fits the training set. The model becomes tailored to the random noise and idiosyncrasies of that specific test sample, so its apparent 95% accuracy is optimistically biased and will not replicate on fresh data. This is sometimes called test set overfitting or 'testing hypotheses on the test set,' and it inflates reported performance relative to true generalization.

Why this answer

The data scientist repeatedly tuned hyperparameters based on test set performance, effectively using the test set as part of the training process. This causes the model to become specialized to the test set's specific patterns and noise, so it fails to generalize to new, unseen data. This phenomenon is known as overfitting to the test set, where the test set no longer provides an unbiased estimate of real-world performance.

Exam trap

The trap here is that candidates confuse overfitting to the training data with overfitting to the test set, failing to recognize that repeatedly evaluating on the test set can cause the model to memorize test set patterns rather than generalize.

How to eliminate wrong answers

Option A is wrong because overfitting to the training data would show high training accuracy but lower test accuracy during evaluation, not a drop only after deployment. Option B is wrong because data leakage would typically inflate test accuracy during tuning, but the scenario describes a drop from 95% to 70% on new data, which is consistent with test set overfitting, not leakage. Option D is wrong because underfitting would result in poor performance on both training and test sets, not a high test accuracy of 95%.

Practice this question →

MCQeasy

What is an endpoint in Azure Machine Learning?

A.A visual dashboard for monitoring model performance

B.A deployed ML model accessible via REST API for making predictions

C.The final training step that produces a saved model file

D.A data storage location for training datasets

AnswerB

In Azure Machine Learning, an endpoint represents a deployed model exposed as a REST API service, providing a scoring URI that client applications invoke with JSON payloads. The endpoint handles authentication, request routing, and returns prediction results, making it the live interface between the model and production applications. This is precisely what an endpoint is: a callable inference service, not a training artifact or monitoring view.

Why this answer

In Azure Machine Learning, an endpoint is a REST API endpoint that exposes a deployed machine learning model for real-time inference. When you deploy a model to an Azure Kubernetes Service (AKS) or Azure Container Instances (ACI) cluster, Azure ML creates a scoring URI that clients can call with HTTP POST requests containing input data, and the endpoint returns predictions. This enables applications to integrate model predictions via standard HTTPS protocol.

Exam trap

The trap here is that candidates confuse the term 'endpoint' with the final step of training or with data storage, because in other Azure services 'endpoint' can refer to a storage endpoint or a training job output, but in Azure ML it specifically means the deployed model's REST API for inference.

How to eliminate wrong answers

Option A is wrong because a visual dashboard for monitoring model performance is called Azure ML Studio's monitoring dashboard or Application Insights integration, not an endpoint. Option C is wrong because the final training step that produces a saved model file is the model registration or training run output, not an endpoint; endpoints are created after deployment. Option D is wrong because a data storage location for training datasets is a datastore (e.g., Azure Blob Storage or Azure Data Lake), not an endpoint.

Practice this question →

MCQmedium

What does 'model accuracy' measure in machine learning classification?

A.How quickly the model makes predictions

B.The proportion of correct predictions out of total predictions

C.How much memory the model uses during inference

D.The number of training examples used to build the model

AnswerB

Accuracy is the classification metric that calculates the proportion of correct predictions out of all predictions made, formally expressed as (true positives + true negatives) divided by the total number of evaluated cases. For instance, if a model correctly labels 90 of 100 test instances, its accuracy is 0.90, or 90%. This metric collapses both false positives and false negatives into a single correctness value, so it summarizes overall performance but must be interpreted cautiously with imbalanced classes.

Why this answer

Model accuracy in classification measures the ratio of correctly predicted instances to the total number of predictions made. It is calculated as (True Positives + True Negatives) / (Total Predictions). This metric is fundamental in evaluating classification models on Azure Machine Learning, where it is reported in the model evaluation metrics.

Exam trap

The trap here is that candidates often confuse model accuracy with performance metrics like speed or resource usage, or assume it relates to training data size, when in fact accuracy strictly measures the proportion of correct predictions.

How to eliminate wrong answers

Option A is wrong because model accuracy does not measure prediction speed; inference latency is measured in milliseconds or seconds, not as a proportion of correct predictions. Option C is wrong because memory usage during inference is a resource consumption metric, not a measure of prediction correctness; Azure monitors memory via metrics like 'Memory Usage' in container instances. Option D is wrong because the number of training examples is a dataset size characteristic, not a model performance metric; accuracy evaluates how well the model generalizes, not how much data was used.

Practice this question →

MCQmedium

A medical research team trains a model to detect a rare disease from lab results. The disease occurs in only 1% of patients. The model predicts 'no disease' for every patient and achieves 99% accuracy. Which metric best reveals that the model is failing to identify actual disease cases?

A.Accuracy

B.Precision

C.Recall

D.F1 score

AnswerC

Recall, also called sensitivity, is computed as true positives divided by the total number of actual positive cases (true positives + false negatives). Here, the model predicts no positive cases, so the true positive count is zero, yielding a recall of 0%. This directly indicates that 100% of actual disease cases are missed, which is the exact failure mode in this screening scenario. For rare disease detection, high recall is essential to ensure that no affected patient is overlooked, even if it means accepting more false positives.

Why this answer

Recall (sensitivity) measures the proportion of actual positive cases correctly identified by the model. With a 99% accuracy but zero true positives (since the model always predicts 'no disease'), recall is 0%, which directly reveals the model's failure to detect any actual disease cases. In Azure Machine Learning, recall is a key metric for imbalanced classification tasks, especially when missing a positive case has severe consequences.

Exam trap

The trap here is that candidates see 99% accuracy and assume the model is performing well, without recognizing that accuracy is meaningless when the class distribution is extremely skewed.

How to eliminate wrong answers

Option A is wrong because accuracy is misleading in highly imbalanced datasets; a model that always predicts the majority class can achieve high accuracy (99%) while completely failing to detect the minority class (disease). Option B is wrong because precision measures the proportion of positive predictions that are actually correct; since the model never predicts positive, precision is undefined (division by zero) and does not reveal the failure to identify actual disease cases. Option D is wrong because the F1 score is the harmonic mean of precision and recall; with recall at 0%, the F1 score is also 0%, but recall alone more directly and intuitively exposes the model's inability to detect any positive cases.

Practice this question →

MCQeasy

A data scientist trains a model to predict house prices using features like number of bedrooms, square footage, and location. The model achieves a mean absolute error (MAE) of $5,000 on the training data but $25,000 on the test data. Which problem is the model most likely experiencing?

A.Underfitting

B.Overfitting

C.Multicollinearity

D.Class imbalance

AnswerB

Overfitting happens when a model fits the training data too closely, capturing random noise and idiosyncrasies that do not generalize to unseen data. For house-price prediction, an overfit model might achieve near-perfect MAE on the training set but produce a much higher MAE on the test set because it has effectively memorized the training examples rather than learning the true price determinants. This large train-test performance gap is the hallmark symptom of overfitting.

Why this answer

The model performs well on training data (MAE $5,000) but poorly on test data (MAE $25,000), which is the classic symptom of overfitting. Overfitting occurs when the model learns noise and specific patterns in the training data too well, failing to generalize to unseen data. In Azure Machine Learning, this can be detected by comparing training vs. validation metrics and is often mitigated using regularization techniques or simplifying the model.

Exam trap

The trap here is that candidates confuse overfitting with underfitting because they see a low training error, but the key is the large gap between training and test error, which is the hallmark of overfitting, not underfitting.

How to eliminate wrong answers

Option A is wrong because underfitting would show poor performance on both training and test data (e.g., high MAE on both), not a large gap between them. Option C is wrong because multicollinearity refers to high correlation between independent variables, which can affect coefficient stability but does not directly cause a large train-test performance gap; it would typically inflate variance in predictions but not produce such a stark contrast. Option D is wrong because class imbalance is a problem for classification tasks (e.g., predicting categories), not for regression tasks like predicting house prices, and it would manifest as poor performance on minority classes, not a train-test MAE gap.

Practice this question →

MCQmedium

A data scientist trains a model to predict the exact number of cars that will cross a bridge each day for maintenance planning. The model uses historical traffic data as input. Which type of machine learning task is this?

A.Classification

B.Regression

C.Clustering

D.Reinforcement learning

AnswerB

Regression predicts a continuous numeric target value, which is exactly what is needed to estimate the exact number of cars. Even though a car count is an integer, regression models such as linear regression or decision tree regression learn a mapping from input features to this numeric quantity and can output a precise estimated value.

Why this answer

The model predicts a continuous numerical value (the exact number of cars) based on historical traffic data. Regression is the correct machine learning task for predicting continuous numeric outcomes, such as counts, prices, or temperatures, making option B correct.

Exam trap

The trap here is that candidates confuse predicting a numeric count with classification, mistakenly thinking 'number of cars' is a category, but regression is required for any continuous numeric output.

How to eliminate wrong answers

Option A is wrong because classification predicts discrete categories or labels (e.g., 'high traffic' or 'low traffic'), not a continuous number. Option C is wrong because clustering groups unlabeled data into clusters based on similarity, without predicting a specific numeric value. Option D is wrong because reinforcement learning involves an agent learning optimal actions through rewards and penalties in an environment, not predicting a numeric output from historical data.

Practice this question →

MCQmedium

What is 'model evaluation' and what metrics are used for different ML task types?

A.Accuracy is the only metric needed for all ML task types

B.Different tasks use different metrics: F1 for classification, RMSE for regression, mAP for detection

C.Model evaluation is only needed before deployment, not after

D.The only reliable evaluation is user feedback after the model is deployed in production

AnswerB

Evaluation metrics are inherently task-specific because each machine learning task optimizes a different notion of error. For classification, especially with imbalanced classes, F1 score balances precision and recall, while AUC evaluates ranking capability; for regression, RMSE quantifies prediction error in the original units, alongside R² for variance explained. For object detection, mAP combines localization and classification accuracy across confidence thresholds, making it the standard metric.

Why this answer

Model evaluation is the process of assessing how well a trained machine learning model performs on unseen data. Different ML task types require different metrics because they measure distinct aspects of performance: for classification tasks, F1-score balances precision and recall; for regression tasks, RMSE (Root Mean Squared Error) quantifies prediction error in the same units as the target; for object detection tasks, mAP (mean Average Precision) evaluates both localization and classification accuracy. Option B correctly identifies these task-specific metrics.

Exam trap

The trap here is that candidates often assume accuracy is a universal metric, but the AI-900 exam specifically tests that different ML tasks (classification, regression, detection) require specialized metrics like F1, RMSE, and mAP to properly evaluate model performance.

How to eliminate wrong answers

Option A is wrong because accuracy is not sufficient for all ML tasks—it fails on imbalanced classification datasets where a model can achieve high accuracy by always predicting the majority class, and it is meaningless for regression or detection tasks. Option C is wrong because model evaluation is an ongoing process that should occur both before deployment (to validate performance on test data) and after deployment (to monitor for data drift, concept drift, and performance degradation in production). Option D is wrong because user feedback is subjective, delayed, and not a quantitative metric; it cannot replace objective evaluation metrics like precision, recall, or RMSE, which provide reproducible and statistically sound performance measurements.

Practice this question →

100

MCQmedium

What is 'regularisation' in machine learning and what problem does it solve?

A.Standardising input features to the same scale before training

B.Adding a penalty to the loss function to discourage overly complex models and reduce overfitting

C.Applying government regulations to ensure AI models comply with data privacy laws

D.Converting irregular training data shapes into a uniform format for the algorithm

AnswerB

Adding a penalty to the loss function is exactly what regularisation accomplishes: the algorithm minimises both the original error and a penalty term such as L1 (sum of absolute weights) or L2 (sum of squared weights). This constraint pushes learned weights toward smaller values, thereby reducing model complexity and mitigating overfitting by improving generalisation to unseen data. The penalty strength is controlled by a hyperparameter (often λ), balancing fit versus simplicity.

Why this answer

Regularisation is a technique used in machine learning to prevent overfitting by adding a penalty term to the loss function. This penalty discourages the model from learning overly complex patterns, such as large or numerous coefficients, which helps the model generalise better to unseen data. In Azure Machine Learning, regularisation parameters like L1 (Lasso) or L2 (Ridge) can be configured in algorithms such as linear regression or neural networks to control model complexity.

Exam trap

The trap here is that candidates confuse regularisation with data preprocessing steps like normalisation or reshaping, because both involve modifying data or model parameters, but regularisation specifically targets overfitting by penalising complexity, not by altering input data format or scale.

How to eliminate wrong answers

Option A is wrong because standardising input features to the same scale is called feature scaling or normalisation, not regularisation; it addresses gradient descent convergence, not overfitting. Option C is wrong because applying government regulations for data privacy is a compliance or governance concern, not a machine learning regularisation technique; it relates to policies like GDPR, not model training. Option D is wrong because converting irregular training data shapes into a uniform format refers to data preprocessing or reshaping, which is unrelated to the penalty-based regularisation that controls model complexity.

Practice this question →

101

MCQhard

A data scientist is evaluating a binary classification model that predicts whether a transaction is fraudulent. The test set contains 1,000 transactions: 990 legitimate and 10 fraudulent. The model's predictions are shown in the confusion matrix below. Confusion matrix: Predicted Legitimate Predicted Fraudulent Actual Legitimate 942 48 Actual Fraudulent 2 8 Which metric should the data scientist prioritize if the business goal is to minimize the number of fraudulent transactions that are missed (false negatives)?

A.Precision

B.Recall

C.Accuracy

D.Specificity

AnswerB

Recall = TP/(TP+FN) measures the proportion of actual fraud cases that are correctly identified. It directly quantifies how many fraudulent transactions the model misses (false negatives). Since the stated goal is to minimize false negatives, recall is the most appropriate evaluation metric because a high recall ensures that few real frauds slip through undetected.

Why this answer

Recall (sensitivity) measures the proportion of actual positives correctly identified, calculated as TP/(TP+FN). With 2 false negatives (missed fraudulent transactions), recall is 8/(8+2)=0.80. Minimizing missed fraud directly corresponds to maximizing recall, making it the correct priority for this business goal.

Exam trap

The trap here is that candidates often pick Accuracy because it seems intuitive, but the severe class imbalance (99% legitimate) makes accuracy a poor metric, while Recall directly addresses the business requirement of minimizing missed fraud.

How to eliminate wrong answers

Option A (Precision) is wrong because precision measures the proportion of predicted positives that are actually positive (TP/(TP+FP)), which focuses on avoiding false alarms, not on catching all fraud. Option C (Accuracy) is wrong because accuracy is (TP+TN)/(total) = (8+942)/1000 = 0.95, which is misleadingly high due to class imbalance (990 legitimate vs 10 fraudulent) and does not reflect the cost of missing fraud. Option D (Specificity) is wrong because specificity measures the proportion of actual negatives correctly identified (TN/(TN+FP)), which is about correctly classifying legitimate transactions, not about minimizing missed fraudulent transactions.

Practice this question →

102

MCQmedium

A data scientist has a dataset containing customer transaction records with features such as age, income, and purchase history, but no labels. The goal is to identify natural groupings of customers for a targeted marketing campaign. Which type of machine learning should be used?

A.Classification

B.Regression

C.Clustering

D.Reinforcement learning

AnswerC

Clustering is an unsupervised learning technique that partitions data points into groups based on feature similarity, requiring no predefined labels. For a dataset of customer transactions, clustering can reveal natural segments, such as purchasing-behavior clusters, by measuring distances among features like amount, frequency, and category. This directly matches the task of discovering hidden groupings without prior knowledge of class membership.

Why this answer

Clustering is the correct choice because the dataset has no labels, and the goal is to discover natural groupings of customers based on feature similarity. Unsupervised learning algorithms like K-Means or DBSCAN partition data into clusters where intra-cluster similarity is high and inter-cluster similarity is low, enabling targeted marketing without pre-existing categories.

Exam trap

The trap here is that candidates confuse clustering with classification because both involve grouping, but classification requires pre-labeled categories while clustering discovers them from unlabeled data.

How to eliminate wrong answers

Option A is wrong because classification requires labeled data to predict discrete class labels, but this dataset has no labels. Option B is wrong because regression predicts continuous numerical values (e.g., income amount) from labeled data, not groupings. Option D is wrong because reinforcement learning involves an agent learning from rewards and punishments in an environment, not from static unlabeled data.

Practice this question →

103

MCQeasy

What is 'Azure Machine Learning Responsible AI dashboard's error analysis'?

A.A log of all Python exceptions and errors that occurred during model training

B.Identifying data subgroups where the model makes disproportionately more errors than average

C.Counting the total number of incorrect predictions across the full test set

D.Reviewing error messages from failed Azure ML pipeline runs to diagnose infrastructure issues

AnswerB

Error analysis surfaces model blind spots by partitioning the dataset into cohorts based on input features, such as age, sex, or region, and comparing prediction accuracy across those cohorts. When a subset shows a disproportionately higher error rate than the overall test set, it indicates a systematic failure that aggregate metrics would obscure. This cohort-based inspection is exactly what Azure ML's error analysis dashboard is designed to reveal, supporting targeted model improvement and fairness review.

Why this answer

Azure Machine Learning Responsible AI dashboard's error analysis is specifically designed to identify data subgroups where the model performs poorly, often revealing bias or systematic failures. It uses a decision tree-based approach to partition the dataset and highlight cohorts with disproportionately high error rates, enabling targeted mitigation. This goes beyond simple aggregate metrics to uncover hidden disparities in model performance.

Exam trap

The trap here is that candidates confuse 'error analysis' with basic error counting or debugging, when the key is its focus on subgroup-level disparity detection, not aggregate or infrastructure errors.

How to eliminate wrong answers

Option A is wrong because error analysis does not log Python exceptions or training errors; it focuses on model prediction errors on test data, not code-level failures. Option C is wrong because counting total incorrect predictions is a basic aggregate metric (e.g., error rate), not the subgroup-level analysis that error analysis provides. Option D is wrong because error analysis evaluates model predictions, not infrastructure or pipeline run errors; diagnosing failed runs is a separate operational concern.

Practice this question →

104

MCQmedium

What is 'imbalanced classification' handling using 'SMOTE'?

A.A technique for collecting more real minority class examples from external data sources

B.Generating synthetic minority class examples by interpolating between existing examples

C.Removing majority class examples until all classes have equal representation

D.Setting model confidence thresholds to classify more examples as the minority class

AnswerB

SMOTE creates synthetic minority-class examples by selecting a minority instance, identifying its k-nearest minority neighbors, and randomly interpolating along the line segment to one of those neighbors. This augments the feature space with plausible variations instead of duplicating identical records, which gives classifiers richer coverage of the rare class and helps ordinary algorithms learn more robust decision boundaries.

Why this answer

SMOTE (Synthetic Minority Over-sampling Technique) is a data augmentation method that creates synthetic examples for the minority class by interpolating between existing minority class instances. It selects a minority example, finds its k-nearest neighbors from the same class, and generates new samples along the line segments connecting the example to those neighbors. This balances the class distribution without duplicating existing data or discarding majority class examples.

Exam trap

The trap here is that candidates confuse SMOTE with undersampling or threshold tuning, but SMOTE is specifically a synthetic oversampling technique that creates new data points, not a method for removing data or adjusting model parameters.

How to eliminate wrong answers

Option A is wrong because SMOTE does not involve collecting real examples from external sources; it generates synthetic data from the existing minority class. Option C is wrong because that describes random undersampling, not SMOTE, which oversamples the minority class rather than removing majority examples. Option D is wrong because adjusting confidence thresholds is a post-training decision boundary technique, not a data-level method like SMOTE for handling imbalanced classification.

Practice this question →

105

MCQmedium

What is a 'compute instance' in Azure Machine Learning?

A.A scalable cluster for running distributed training jobs across many nodes

B.A managed cloud workstation for interactive ML development with pre-installed tools

C.A virtual machine that automatically scales to run batch predictions

D.A serverless execution environment for ML inference requests

AnswerB

A compute instance is a managed, single-node cloud workstation pre-configured with Azure Machine Learning Studio, Python environments, notebooks, and common development tools such as VS Code. Data scientists use it interactively to write code, explore data, prototype models, and run experiments without manually provisioning or configuring a virtual machine. This is exactly the definition of an Azure Machine Learning compute instance, so this is the correct answer.

Why this answer

A compute instance in Azure Machine Learning is a fully managed cloud workstation that provides a pre-configured environment with popular ML tools like Jupyter Notebooks, TensorFlow, and PyTorch. It is designed for interactive development, allowing data scientists to train and experiment with models without managing infrastructure.

Exam trap

The trap here is that candidates confuse 'compute instance' with 'compute cluster' because both are compute targets, but the instance is for single-user interactive work while the cluster is for multi-node distributed jobs.

How to eliminate wrong answers

Option A is wrong because a scalable cluster for running distributed training jobs across many nodes describes an Azure Machine Learning compute cluster, not a compute instance. Option C is wrong because a virtual machine that automatically scales to run batch predictions describes an Azure Machine Learning inference cluster or a managed online endpoint, not a compute instance. Option D is wrong because a serverless execution environment for ML inference requests describes Azure Machine Learning serverless inference endpoints or Azure Functions, not a compute instance.

Practice this question →

106

MCQmedium

A data scientist is training a regression model to predict house prices using features like square footage, number of bedrooms, and location. After evaluating the model on a test set, the data scientist wants to select a metric that measures the average magnitude of prediction errors in the same units as the target variable (price). Which evaluation metric should the data scientist use?

A.Root Mean Squared Error (RMSE)

B.Accuracy

C.F1 Score

D.Precision

AnswerA

Root Mean Squared Error (RMSE) is the square root of the average of the squared differences between predicted and actual house prices. Because it operates in the same units as the target variable and heavily penalizes large errors, it directly quantifies prediction accuracy for continuous regression outputs.

Why this answer

Root Mean Squared Error (RMSE) is the correct metric because it measures the average magnitude of prediction errors in the same units as the target variable (price). RMSE is computed as the square root of the average squared differences between predicted and actual values, which brings the error metric back to the original unit (e.g., dollars), making it directly interpretable for regression tasks like house price prediction.

Exam trap

The trap here is that candidates often confuse regression metrics with classification metrics, mistakenly selecting Accuracy or F1 Score because they are familiar from other contexts, without recognizing that the question explicitly asks for a metric measuring error magnitude in the same units as the target variable, which only RMSE (or MAE) satisfies.

Why the other options are wrong

Accuracy is a classification metric that measures the proportion of correct predictions, not the magnitude of errors in regression. It does not provide error magnitude in the same units as the target variable.

F1 Score is a classification metric that balances precision and recall, not suitable for regression tasks like predicting house prices.

Precision is a classification metric that measures the proportion of true positive predictions among all positive predictions, not applicable to regression tasks like predicting house prices.

Practice this question →

107

MCQmedium

What is 'online learning' (incremental learning) in machine learning?

A.Training ML models through an online learning management system

B.Continuously updating model weights on new data as it arrives rather than batch retraining

C.Requiring an internet connection during model training for cloud compute access

D.A training approach where users can interact with and correct the model in real time

AnswerB

This is the precise definition of online learning, also called incremental learning: instead of periodically retraining on the entire historical dataset, the model's weights are adjusted continuously with each new data point or small batch using an update rule like gradient descent. This approach adapts quickly to non-stationary data distributions and is ideal for real-time streams such as fraud detection, sensor telemetry, or clickstream feedback. However, a critical drawback is catastrophic forgetting, where the model may overwrite previously acquired patterns when the incoming data distribution shifts sharply.

Why this answer

Online learning (incremental learning) is a machine learning technique where the model is updated continuously as new data arrives, rather than retraining from scratch on the entire dataset. This is essential for scenarios with streaming data or when retraining on all historical data is computationally prohibitive. In Azure, this is supported by services like Azure Stream Analytics and Azure Machine Learning's online endpoints, which can update model weights incrementally.

Exam trap

The trap here is confusing 'online learning' with 'requiring an internet connection' (Option C) or with 'interactive human correction' (Option D), when the term specifically refers to incremental data ingestion and model weight updates.

How to eliminate wrong answers

Option A is wrong because it describes a learning management system (LMS) for human education, not a machine learning training paradigm. Option C is wrong because while cloud compute may be used, online learning does not require an internet connection; it refers to incremental data processing, not network connectivity. Option D is wrong because it describes interactive or active learning where humans correct the model, which is a different concept from automated incremental weight updates based on new data.

Practice this question →

108

MCQmedium

A data scientist is training a model to predict whether a patient has a rare disease (1% prevalence). The model predicts 'no disease' for all patients and achieves 99% accuracy, but fails to identify any actual cases. Which metric would best reveal this failure?

A.Precision

B.Recall

C.F1 score

D.Mean absolute error

AnswerB

Recall (sensitivity) measures the proportion of actual positive cases correctly identified, calculated as TP/(TP+FN). With zero positive predictions, TP = 0 and every actual disease case becomes a false negative, so FN equals the total number of positives, yielding a recall of 0%. This directly exposes that the model misses all patients with the disease, which is precisely the failure mode in question.

Why this answer

Recall (sensitivity) measures the proportion of actual positive cases correctly identified. With 1% disease prevalence and a model that predicts 'no disease' for all patients, recall is 0% because zero true positives are found. Accuracy (99%) is misleading here because the model fails to detect any rare disease cases, and recall directly exposes this failure.

Exam trap

The trap here is that candidates see 99% accuracy and assume the model is performing well, failing to recognize that accuracy is a poor metric for imbalanced datasets and that recall specifically measures the model's ability to catch rare positive cases.

How to eliminate wrong answers

Option A is wrong because precision measures the proportion of positive predictions that are correct; since the model never predicts positive, precision is undefined or 0, but precision does not directly reveal the failure to find actual cases. Option C is wrong because the F1 score is the harmonic mean of precision and recall; with recall at 0%, the F1 score is also 0, but it does not isolate the failure as clearly as recall does. Option D is wrong because mean absolute error (MAE) is a regression metric used for continuous values, not for binary classification tasks like disease prediction.

Practice this question →

109

MCQeasy

What is 'Azure Machine Learning compute' and what types are available?

A.The mathematical computations performed by the model during training

B.The managed cloud infrastructure (VMs, clusters) used to run ML training and inference workloads

C.The number of floating-point operations a model performs per second

D.A billing calculator that estimates the cost of running machine learning workloads

AnswerB

Azure ML compute is the managed cloud infrastructure used to run ML training and inference workloads. It encompasses compute instances for interactive development, compute clusters for scalable parallel training with auto-scaling and job scheduling, and inference clusters for deploying models as endpoints. These compute targets abstract away the need to manage raw VMs, providing integrated security, identity, and integration with Azure ML pipelines.

Why this answer

Azure Machine Learning compute is a managed cloud infrastructure that provides on-demand virtual machines (VMs) and clusters for running machine learning training and inference workloads. It abstracts away the underlying hardware management, allowing you to dynamically scale compute resources up or down based on job requirements, and supports both CPU and GPU instances for different model types.

Exam trap

The trap here is confusing the abstract concept of 'compute' (the infrastructure) with the mathematical computations or performance metrics, leading candidates to pick A or C instead of recognizing it as a managed cloud resource.

How to eliminate wrong answers

Option A is wrong because it describes the mathematical operations (e.g., matrix multiplications) performed during model training, which is a computational process, not the infrastructure that runs it. Option C is wrong because it refers to FLOPS (floating-point operations per second), a performance metric for measuring computational throughput, not the managed compute service itself. Option D is wrong because it describes the Azure Pricing Calculator or TCO calculator, which estimates costs but does not execute ML workloads.

Practice this question →

110

MCQmedium

A data scientist is training a regression model to predict house prices. The data scientist wants to evaluate the model using a metric that penalizes large prediction errors significantly more than small errors. Which evaluation metric should the data scientist choose?

A.Mean Absolute Error (MAE)

B.Root Mean Squared Error (RMSE)

C.R-squared (R²)

D.Mean Absolute Percentage Error (MAPE)

AnswerB

RMSE squares the errors before averaging and then takes the square root. The squaring step causes larger errors to have a disproportionately higher impact on the metric, making it sensitive to outliers and large deviations.

Why this answer

Root Mean Squared Error (RMSE) is the correct choice because it squares the residuals before averaging, which heavily penalizes large prediction errors (outliers) more than small errors. This aligns with the requirement to penalize large errors significantly more than small ones, as the squaring operation amplifies the impact of larger deviations.

Exam trap

The trap here is that candidates often confuse MAE with RMSE, thinking both penalize errors equally, but the squaring operation in RMSE is the key differentiator that makes it penalize large errors disproportionately.

Why the other options are wrong

MAE treats all errors equally, so it does not penalize large errors more than small errors, which is the requirement in the question.

R-squared measures the proportion of variance explained by the model, not the magnitude of prediction errors. It does not penalize large errors more than small errors, as it is based on squared deviations but is scale-invariant and not directly an error metric.

MAPE does not penalize large errors significantly more than small errors; it treats errors proportionally to the actual values, and large errors can be masked by small actual values. The question specifically requires a metric that heavily penalizes large errors, which RMSE does via squaring.

Practice this question →

111

MCQmedium

Which type of machine learning uses labeled training data where the correct output is provided for each input?

A.Unsupervised learning

B.Reinforcement learning

C.Supervised learning

D.Transfer learning

AnswerC

Supervised learning trains models on labelled datasets, where each input is paired with its known correct output, enabling the algorithm to learn the mapping between them. This directly satisfies the stem's requirement for labelled training data with provided outputs, distinguishing it from unsupervised learning, which uses unlabelled data.

Why this answer

Supervised learning is the correct answer because it explicitly uses labeled training data where each input example is paired with the correct output label. The algorithm learns to map inputs to outputs by minimizing the error between its predictions and the provided labels, enabling tasks like classification and regression.

Exam trap

The trap here is that candidates often confuse 'supervised learning' with 'reinforcement learning' because both involve feedback, but reinforcement learning uses delayed rewards from actions rather than direct labeled examples.

How to eliminate wrong answers

Option A is wrong because unsupervised learning uses unlabeled data and finds hidden patterns or groupings without any correct output provided. Option B is wrong because reinforcement learning learns through trial-and-error interactions with an environment using rewards and penalties, not from pre-labeled input-output pairs. Option D is wrong because transfer learning is a technique that reuses a pre-trained model on a new but related task, not a distinct learning paradigm that uses labeled training data directly.

Practice this question →

112

MCQeasy

A data scientist is building a model to predict the exact temperature in degrees Celsius based on humidity and atmospheric pressure. The model will output a single numeric value for each input. Which type of machine learning task is this?

A.Classification

B.Regression

C.Clustering

D.Object detection

AnswerB

Regression predicts a continuous numeric output, here an exact temperature value in degrees Celsius from humidity and pressure inputs. Classification would output discrete labels or categories instead, so regression is the task type that fits this continuous prediction.

Why this answer

This is a regression task because the goal is to predict a continuous numeric value (temperature in degrees Celsius) from input features (humidity and atmospheric pressure). Regression models output a real number, unlike classification which predicts discrete categories. In Azure Machine Learning, regression algorithms like Linear Regression or Decision Forest Regression are used for such tasks.

Exam trap

The trap here is that candidates may confuse predicting a numeric value with classification, but classification outputs discrete labels (e.g., 'high temperature' vs 'low temperature'), not a precise continuous number like degrees Celsius.

Why the other options are wrong

Classification predicts discrete labels or categories, not continuous numeric values. The question asks for a single numeric temperature value, which is a regression task.

Clustering groups data into clusters without labeled outputs, but this question requires predicting a continuous numeric value (temperature) from inputs, which is a regression task.

Object detection is used to identify and locate objects within images or videos, not to predict a continuous numeric value like temperature from numerical inputs.

Practice this question →

113

MCQmedium

What is 'transfer learning' and how is it different from training from scratch?

A.Transfer learning and training from scratch produce identical results

B.Transfer learning fine-tunes a pre-trained model on a new task — requiring far less data and compute than training from scratch

C.Transfer learning copies model weights between Azure subscriptions

D.Transfer learning is used only when the original training data is unavailable

AnswerB

Transfer learning takes a model already trained on a broad, large-scale dataset and fine-tunes it on a smaller, task-specific dataset, often with a low learning rate. The early layers retain generic features, while later layers adapt to the new task, so the model needs only a fraction of the data and compute that training from scratch would require. This is why pre-trained models can be adapted to custom scenarios with relatively few labeled examples.

Why this answer

Transfer learning starts with a model already trained on a large dataset (e.g., ImageNet) and fine-tunes it on a smaller, task-specific dataset. This approach requires significantly less data and computational resources compared to training from scratch, where all model weights are randomly initialized and learned from the ground up. It is especially effective when the new task is similar to the original training task, allowing the pre-trained features to be reused.

Exam trap

The trap here is that candidates may confuse transfer learning with simply reusing a model without any retraining, or think it only applies when original data is missing, rather than understanding it as a resource-efficient fine-tuning strategy.

How to eliminate wrong answers

Option A is wrong because transfer learning and training from scratch do not produce identical results; transfer learning typically converges faster and may achieve higher accuracy with limited data, while training from scratch requires more data and compute to reach comparable performance. Option C is wrong because transfer learning is a machine learning technique involving model weights, not a mechanism for copying model weights between Azure subscriptions; Azure subscriptions are unrelated to the concept. Option D is wrong because transfer learning can be used even when original training data is available; it is chosen to save resources and improve performance, not solely due to data unavailability.

Practice this question →

114

MCQmedium

What is 'k-fold cross-validation' specifically and how is k=10 different from k=5?

A.k=10 always produces a better model than k=5 because it uses more training data

B.k=10 provides more reliable performance estimates at 2x the compute cost vs k=5

C.k=5 and k=10 produce identical results because the total data is the same

D.k=10 requires 10 times more labelled data than k=5

AnswerB

k-fold cross-validation partitions data into k subsets, training on k-1 and validating on the held-out fold. Raising k from 5 to 10 halves each validation set's size, giving lower-variance estimates while doubling the number of training runs.

Why this answer

k-fold cross-validation splits the dataset into k equal folds, training on k-1 folds and validating on the remaining fold, repeating this process k times. With k=10, each model is trained on 90% of the data and validated on 10%, while k=5 uses 80% for training and 20% for validation. The key difference is that k=10 yields a performance estimate with lower variance (more reliable) because it averages over more folds, but it requires approximately twice the computational cost (10 training runs vs. 5).

Exam trap

The trap here is confusing model performance improvement with estimate reliability; candidates often think more folds always yield a better model, but cross-validation is about evaluating performance, not training the final model.

How to eliminate wrong answers

Option A is wrong because k=10 does not always produce a better model; it provides a more reliable estimate of model performance, but the actual model quality depends on the algorithm and data, not the cross-validation fold count. Option C is wrong because k=5 and k=10 produce different results due to different training/validation splits and variance in estimates; they are not identical. Option D is wrong because k-fold cross-validation does not require more labelled data; it uses the same dataset, just partitioned differently.

Practice this question →

115

MCQhard

A data scientist trains a binary classification model to detect fraudulent transactions. The dataset contains only 2% fraudulent transactions. The model achieves 98% overall accuracy, but it fails to detect any fraudulent transactions, classifying all transactions as legitimate. Which metric would most clearly reveal this failure?

A.Precision

B.Recall

C.F1 score

D.Specificity

AnswerB

Recall, or true positive rate, is TP/(TP+FN). Since the model never predicts fraud, TP=0 while FN equals the total number of actual fraud cases, making recall exactly zero. This metric measures the model's ability to find positive examples, so it clearly and unambiguously exposes the model's failure to detect any fraudulent transactions.

Why this answer

Recall (also known as sensitivity or true positive rate) measures the proportion of actual positive cases (fraudulent transactions) that were correctly identified by the model. In this scenario, the model classifies all transactions as legitimate, so it detects zero fraudulent transactions, yielding a recall of 0%. Despite 98% overall accuracy, the recall metric clearly exposes the model's complete failure to identify any fraud.

Exam trap

The trap here is that candidates often assume high overall accuracy (98%) implies good model performance, failing to recognize that accuracy is a poor metric for imbalanced datasets and that recall is the metric that directly exposes the model's inability to detect the minority class.

How to eliminate wrong answers

Option A is wrong because precision measures the proportion of predicted positive cases that are actually positive; if the model predicts no positives, precision is undefined (division by zero) or 0/0, which does not clearly reveal the failure to detect fraud. Option C is wrong because the F1 score is the harmonic mean of precision and recall; with recall at 0%, the F1 score will also be 0%, but it does not directly highlight the failure as intuitively as recall alone. Option D is wrong because specificity measures the proportion of actual negative cases (legitimate transactions) correctly identified; the model correctly classifies all legitimate transactions, so specificity would be 100%, masking the failure to detect fraud.

Practice this question →

116

MCQhard

A data scientist is training a binary classification model to detect rare equipment failures from sensor data. The dataset contains 99.5% normal operation readings and only 0.5% failure readings. The model currently predicts all readings as 'normal' and achieves 99.5% accuracy on the test set. The business requires the model to identify at least 80% of actual failures. Which data-level technique should the data scientist use to most directly address the class imbalance?

A.Oversample the minority class (failure examples)

B.Undersample the majority class (normal examples)

C.Use precision as the optimization metric

D.Reduce the complexity of the model

AnswerA

Oversampling the minority class replicates or synthetically generates additional failure examples so the training set has a more balanced class distribution. This directly gives the model more opportunities to learn the boundary of rare failure patterns, reducing the tendency to simply predict the majority class. Techniques like SMOTE create interpolated samples rather than exact duplicates, which generally improves generalization on the minority class.

Why this answer

Oversampling the minority class (failure examples) directly addresses the severe class imbalance by creating synthetic copies or duplicates of the rare failure instances. This balances the training dataset, allowing the model to learn patterns associated with failures rather than always predicting the majority class. With a balanced dataset, the model can be trained to meet the business requirement of identifying at least 80% of actual failures, even though overall accuracy may decrease.

Exam trap

The trap here is that candidates may think high accuracy (99.5%) is always good, but in imbalanced datasets, accuracy is misleading; the question tests whether you recognize that data-level techniques like oversampling are needed to force the model to learn the minority class, not just optimize metrics or simplify the model.

How to eliminate wrong answers

Option B is wrong because undersampling the majority class discards the vast majority of normal operation data, which can lead to loss of valuable information and reduced model generalization, especially when the majority class is 99.5% of the data. Option C is wrong because using precision as the optimization metric does not directly address the class imbalance at the data level; it is a model evaluation metric that can be used after rebalancing, but it does not change the underlying skewed distribution. Option D is wrong because reducing model complexity does not fix the class imbalance; it may help prevent overfitting but will not enable the model to learn from the rare failure class when it is vastly underrepresented in the training data.

Practice this question →

117

MCQeasy

What is feature engineering in machine learning?

A.Designing the hardware chips for running ML models

B.Selecting, transforming, and creating input variables from raw data to improve model performance

C.Selecting which neural network layers to include in a model

D.Writing code to deploy ML models as REST APIs

AnswerB

Feature engineering is the deliberate process of selecting, transforming, and creating input variables (features) from raw data to make patterns more accessible to a machine learning algorithm. It includes techniques such as one-hot encoding categorical variables, normalizing numeric ranges, binning continuous values, extracting date/time components, and constructing interaction terms—all aimed at improving model accuracy, convergence speed, and generalization. This is the correct definition because it emphasizes the data representation transformation that precedes model training, distinct from hardware, deployment, or architecture choices.

Why this answer

Feature engineering is the process of selecting, transforming, and creating input variables (features) from raw data to improve the performance of machine learning models. This step is critical because the quality and relevance of features directly impact a model's ability to learn patterns and generalize to new data. In Azure Machine Learning, feature engineering is often performed using tools like the 'Feature Engineering' step in automated ML or custom Python scripts with libraries such as pandas and scikit-learn.

Exam trap

The trap here is that candidates confuse feature engineering with model architecture design (Option C) or deployment (Option D), because all three are part of the ML lifecycle but serve distinct purposes—feature engineering focuses solely on input data transformation, not on model structure or serving.

How to eliminate wrong answers

Option A is wrong because designing hardware chips for running ML models is a hardware engineering task, not a data preprocessing or feature creation activity; it relates to specialized processors like GPUs or FPGAs, not to feature engineering. Option C is wrong because selecting neural network layers is part of model architecture design (e.g., choosing the number of layers in a deep learning model), which occurs after feature engineering and focuses on model structure, not input variable manipulation. Option D is wrong because writing code to deploy ML models as REST APIs is a deployment and MLOps activity, typically using tools like Azure Kubernetes Service or Azure Functions, and has nothing to do with transforming raw data into features.

Practice this question →

118

MCQmedium

What is the F1 score in machine learning evaluation?

A.The first evaluation metric calculated before training a model

B.The harmonic mean of precision and recall that balances both metrics

C.The proportion of predictions correct on the test set

D.A measure of how fast the model produces predictions

AnswerB

F1 is the harmonic mean of precision and recall, defined mathematically as F1 = 2 * (precision * recall) / (precision + recall). The harmonic mean is severe when precision and recall are unbalanced: a model with high precision and low recall receives a low F1 because both false positives and false negatives are penalized through the combination. This makes F1 a balanced measure of a model's reliability in detecting the positive class, particularly for imbalanced datasets.

Why this answer

The F1 score is defined as the harmonic mean of precision and recall, calculated as 2 * (precision * recall) / (precision + recall). This metric provides a single score that balances both false positives and false negatives, making it especially useful when classes are imbalanced. In Azure Machine Learning, the F1 score is a standard evaluation metric for classification models, reported in automated ML runs and designer modules.

Exam trap

The trap here is that candidates confuse the F1 score with accuracy (Option C) because both are single-number metrics, but the F1 score specifically addresses the trade-off between precision and recall, not just overall correctness.

How to eliminate wrong answers

Option A is wrong because the F1 score is an evaluation metric computed after model training and prediction, not before training; metrics like accuracy or loss are not calculated prior to training. Option C is wrong because it describes accuracy (the proportion of correct predictions), not the F1 score, which specifically balances precision and recall. Option D is wrong because it describes inference speed or latency, which is a performance metric unrelated to the statistical evaluation of classification quality.

Practice this question →

119

MCQmedium

What is overfitting in machine learning?

A.When a model performs well on training data but poorly on new, unseen data

B.When a model is trained with too little data

C.When a model takes too long to train

D.When a model performs poorly on both training and test data

AnswerA

Overfitting occurs when a model memorises training data, including noise, so training performance is high while generalisation to new, unseen data is poor. This gap between training and validation performance is the defining characteristic of the condition.

Why this answer

Overfitting occurs when a machine learning model learns the training data too well, including its noise and outliers, resulting in high accuracy on training data but poor generalization to new, unseen data. This is a fundamental concept in ML because the goal is to create models that perform well on real-world data, not just the data they were trained on. In Azure Machine Learning, techniques like regularization, cross-validation, and early stopping are used to detect and mitigate overfitting.

Exam trap

The trap here is that candidates confuse overfitting with underfitting (Option D) or mistakenly think overfitting is caused solely by insufficient data (Option B), when in fact overfitting is about the model's inability to generalize due to excessive complexity or noise memorization.

How to eliminate wrong answers

Option B is wrong because training with too little data can lead to underfitting (high bias) or high variance, but overfitting is specifically about the model memorizing the training data, not the quantity of data alone. Option C is wrong because training time is a performance metric, not a definition of overfitting; a model can overfit quickly or slowly depending on complexity and data size. Option D is wrong because poor performance on both training and test data describes underfitting (high bias), where the model is too simple to capture underlying patterns, not overfitting.

Practice this question →

120

MCQeasy

A data scientist wants to group customers into segments based on purchasing behavior without using any labeled examples. Which type of machine learning is this?

A.Supervised learning

B.Unsupervised learning

C.Reinforcement learning

D.Semi-supervised learning

AnswerB

Unsupervised learning is the correct approach because it identifies natural groupings within data without requiring pre-assigned labels. Algorithms such as k-means, DBSCAN, or hierarchical clustering operate on feature vectors (e.g., purchase frequency, average basket size, product categories) and group customers by similarity—minimizing intra-cluster distance while maximizing inter-cluster separation. Since the data scientist wants to discover segments solely from purchasing behavior and has no known segment labels, this exploratory clustering problem is a textbook use case for unsupervised learning.

Why this answer

Unsupervised learning is the correct choice because the data scientist has no labeled examples and wants to discover hidden patterns or groupings in the data. Clustering algorithms, such as K-Means or DBSCAN, are used to segment customers based solely on their purchasing behavior features, without any predefined categories.

Exam trap

The trap here is that candidates may confuse 'no labeled examples' with semi-supervised learning, but the key distinction is that semi-supervised learning still requires at least some labeled data, while this scenario uses none.

Why the other options are wrong

Supervised learning requires labeled data to train a model, but the question explicitly states no labeled examples are used.

Reinforcement learning involves an agent learning from rewards and punishments through interaction with an environment, not from unlabeled data for grouping customers.

Semi-supervised learning uses a small amount of labeled data alongside unlabeled data, but the question explicitly states 'without using any labeled examples', making unsupervised learning the correct choice.

Practice this question →

121

MCQmedium

A data scientist trains a model on historical data and achieves high accuracy on both the training set and a held-out test set. However, when the model is deployed in production, it performs poorly on new, unseen data. Which issue is most likely the cause?

A.Overfitting

B.Underfitting

C.Data leakage

D.Concept drift

AnswerC

Data leakage is the correct diagnosis: it happens when the training features contain information that would not be available at inference time, such as the target itself, a future value, or a post-outcome field. The model exploits this hidden shortcut, achieving very high accuracy on both training and test splits during offline evaluation, but in real-world deployment the leaked signal is absent, causing immediate and drastic performance collapse. This pattern of artificially perfect historical evaluation followed by severe production failure is the classic signature of data leakage.

Why this answer

Data leakage occurs when information from outside the training dataset is inadvertently used to train the model, causing it to learn patterns that do not generalize to new data. In this scenario, the high accuracy on both training and test sets but poor production performance indicates that the test set was contaminated with information from the future or from the target variable, making the model appear accurate during validation but fail in real-world deployment.

Exam trap

The trap here is that candidates confuse high accuracy on both training and test sets with overfitting, but the key differentiator is that overfitting would show a significant gap between training and test accuracy, whereas data leakage produces deceptively high accuracy on both sets.

How to eliminate wrong answers

Option A is wrong because overfitting would show high training accuracy but low test accuracy, not high accuracy on both sets. Option B is wrong because underfitting would result in poor performance on both training and test sets, not high accuracy. Option D is wrong because concept drift refers to a change in the underlying data distribution over time after deployment, not a static mismatch between training and production data at the time of deployment.

Practice this question →

122

MCQmedium

A hospital has a dataset with historical patient records, each labeled as either 'readmitted within 30 days' or 'not readmitted'. The hospital wants to train a model to predict which current patients are likely to be readmitted. Which type of machine learning task is this?

A.Supervised regression

B.Supervised classification

C.Unsupervised clustering

D.Reinforcement learning

AnswerB

Supervised classification is appropriate because each patient record includes a known outcome label: either 'readmitted' or 'not readmitted' within a specified timeframe. The algorithm learns a decision boundary from historical features (demographics, diagnoses, lab values) to predict the categorical target for new patients. Common algorithms include logistic regression, random forests, and gradient-boosted trees, and the model is evaluated with metrics such as accuracy, precision, recall, and AUC-ROC.

Why this answer

This is a supervised classification task because the dataset contains labeled historical patient records (readmitted or not readmitted), and the goal is to predict a discrete category (binary outcome) for new patients. In Azure Machine Learning, this would use a classification algorithm like logistic regression or decision tree to assign each patient to one of the two classes.

Exam trap

The trap here is that candidates confuse regression with classification when the target variable is a binary outcome, mistakenly thinking 'readmitted or not' is a numeric value rather than a categorical label.

Why the other options are wrong

The task involves predicting a categorical label ('readmitted' or 'not'), not a continuous numeric value. Regression is for predicting continuous outcomes, so it is incorrect here.

The dataset has labeled outcomes (readmitted or not), making it a supervised learning problem, not unsupervised. Clustering is used when no labels exist.

Reinforcement learning involves an agent learning through trial-and-error interactions with an environment to maximize cumulative reward, which does not apply to predicting readmission from labeled historical data.

Practice this question →

123

MCQmedium

What is 'Azure Machine Learning's Responsible AI dashboard' and what does it include?

A.A compliance checklist confirming a model meets Microsoft's responsible AI certification requirements

B.A unified tool for error analysis, interpretability, fairness, data exploration, and causal inference

C.A monitoring dashboard showing responsible AI policy violations in production

D.A report auto-generated and submitted to regulators when a model is deployed

AnswerB

The Responsible AI dashboard is a unified interface in Azure Machine Learning that integrates six complementary lenses: error analysis, interpretability, fairness, data exploration, counterfactual what-if, and causal inference. This consolidation lets developers examine where a model fails, why specific predictions occur, how outcomes vary across demographic groups, and what interventions might change results—all from a single tool. It is specifically designed for iterative, diagnostic model evaluation rather than governance or compliance.

Why this answer

The Responsible AI dashboard in Azure Machine Learning is a unified, integrated tool that combines multiple components for building and evaluating AI systems responsibly. It includes error analysis, model interpretability, fairness assessment, data exploration, and causal inference capabilities, all accessible through a single interface. This dashboard helps data scientists and developers understand model behavior, identify potential biases, and make informed decisions throughout the ML lifecycle.

Exam trap

The trap here is that candidates confuse the Responsible AI dashboard with a compliance or monitoring tool, when in fact it is an interactive analysis and debugging suite for understanding model behavior before deployment.

How to eliminate wrong answers

Option A is wrong because the Responsible AI dashboard is not a compliance checklist or certification tool; it does not confirm that a model meets Microsoft's responsible AI certification requirements, as no such formal certification exists within Azure ML. Option C is wrong because the dashboard is designed for pre-deployment analysis and model understanding, not for monitoring production policy violations; production monitoring is handled by separate tools like Azure Monitor and Model Data Collector. Option D is wrong because the dashboard does not auto-generate or submit reports to regulators; it is an interactive tool for internal analysis, not a regulatory compliance reporting mechanism.

Practice this question →

124

MCQeasy

What is 'automated machine learning' (AutoML) in Azure Machine Learning?

A.A system that automatically retrains models on a fixed daily schedule

B.Automatically iterating through algorithms and hyperparameters to find the best model for a dataset

C.Automatically labelling training data using existing model predictions

D.A robot that physically connects GPU hardware for distributed training

AnswerB

AutoML, or Automated Machine Learning, systematically explores a defined search space of candidate algorithms (e.g., linear regression, tree ensembles, deep networks) and their hyperparameter configurations. It uses techniques like Bayesian optimization, random search, or early termination to evaluate many candidates against a validation metric and automatically selects the best-performing model. This directly replaces the manual, iterative trial-and-error performed by a data scientist, which is the central capability that defines AutoML.

Why this answer

Automated machine learning (AutoML) in Azure Machine Learning automates the process of selecting the best machine learning algorithm and tuning its hyperparameters for a given dataset. It iterates through multiple combinations of algorithms and hyperparameter values, evaluating each model's performance to identify the optimal solution without manual intervention. This is why option B is correct.

Exam trap

The trap here is that candidates confuse AutoML with simple scheduled retraining (option A) or with automated data labeling (option C), but the core definition of AutoML is specifically about automating the algorithm selection and hyperparameter tuning process.

How to eliminate wrong answers

Option A is wrong because AutoML does not simply retrain models on a fixed daily schedule; that describes a scheduled retraining pipeline, not the automated algorithm and hyperparameter search process. Option C is wrong because automatically labeling training data using existing model predictions is known as 'pseudo-labeling' or 'self-training', not AutoML. Option D is wrong because AutoML is a software-based optimization process, not a physical robot that connects GPU hardware for distributed training.

Practice this question →

125

MCQhard

What is 'causal inference' and how does it differ from correlation-based machine learning?

A.Causal inference uses larger training datasets; correlation-based ML uses smaller ones

B.Causal inference determines whether X actually causes Y; ML finds correlations that predict outcomes

C.Causal inference is exclusively used in medical research; ML is used in business applications

D.ML models always establish causal relationships; causal inference is needed only when data quality is poor

AnswerB

This is correct. Causal inference goes beyond observed patterns to answer questions like 'if we intervene and set X, does Y change?' using tools such as randomized experiments, propensity scores, or structural causal models. Standard ML typically optimizes a loss function to find correlations in historical data that predict outcomes, but it does not model interventions or counterfactuals, so it cannot distinguish a true cause from a spurious correlation.

Why this answer

Causal inference specifically aims to determine whether a change in variable X directly causes a change in variable Y, often through controlled experiments or techniques like do-calculus. In contrast, correlation-based machine learning identifies statistical patterns and associations between variables to make predictions, but does not establish a cause-and-effect relationship. This distinction is fundamental in Azure Machine Learning when choosing between predictive modeling (e.g., regression) and causal analysis (e.g., using the DoWhy library).

Exam trap

The trap here is that candidates often confuse correlation with causation, assuming that a strong predictive relationship in ML implies a causal link, when in fact causal inference requires additional experimental or quasi-experimental methods to establish causality.

How to eliminate wrong answers

Option A is wrong because the size of the training dataset is not a defining difference between causal inference and correlation-based ML; both can use large or small datasets depending on the problem. Option C is wrong because causal inference is not exclusively used in medical research; it is applied in economics, social sciences, and business (e.g., A/B testing on Azure). Option D is wrong because ML models do not always establish causal relationships; they typically find correlations, and causal inference is needed when you want to understand the effect of an intervention, not just when data quality is poor.

Practice this question →

126

MCQmedium

What is 'batch inference' vs 'real-time inference' in Azure Machine Learning?

A.Batch inference is more accurate; real-time is faster but less accurate

B.Real-time processes individual requests immediately; batch processes large datasets at scheduled intervals

C.Batch requires GPU compute; real-time uses CPU only

D.Real-time inference is only available in Azure; batch works on-premises too

AnswerB

That is correct: real-time inference (also called online inference) responds to each individual request with a prediction immediately, typically via a REST API, enabling interactive applications like a chatbot or fraud detector. Batch inference, on the other hand, processes a large dataset asynchronously in a scheduled or otherwise triggered job, producing predictions for many records collected together, which suits periodic scoring like daily customer churn analyses. This distinction in latency and workload shape—not accuracy or hardware—is the primary reason you choose one pattern over the other.

Why this answer

Batch inference processes large datasets asynchronously at scheduled intervals, making it suitable for offline or periodic predictions, while real-time inference handles individual requests immediately with low latency for interactive applications. Azure Machine Learning supports both: real-time endpoints for synchronous scoring and batch endpoints for asynchronous, high-throughput processing.

Exam trap

The trap here is that candidates confuse 'batch' with 'less accurate' or 'real-time' with 'GPU-only', when in fact the core distinction is synchronous vs asynchronous processing, not performance or hardware constraints.

How to eliminate wrong answers

Option A is wrong because accuracy is not inherently tied to inference mode; both batch and real-time inference use the same trained model, so accuracy is identical. Option C is wrong because neither batch nor real-time inference is restricted to a specific compute type; both can use CPU or GPU depending on the model and workload requirements. Option D is wrong because real-time inference is not exclusive to Azure; it can be deployed on-premises or in other cloud environments, and batch inference also works on-premises via Azure Arc or local deployments.

Practice this question →

127

MCQeasy

What is 'Azure Machine Learning notebooks' and who typically uses them?

A.Digital note-taking applications for recording meeting minutes during ML project planning

B.Interactive Jupyter notebook environments for data exploration and model prototyping by data scientists

C.Automated logging notebooks that record all model training metrics without code

D.Read-only document viewers for reviewing completed ML experiment results

AnswerB

Azure Machine Learning notebooks are cloud-hosted Jupyter environments that run on compute instances, giving data scientists interactive access to Python or R kernels. These notebooks support inline code execution, data visualizations via matplotlib, and iterative experimentation, enabling rapid prototyping of models directly against Azure datasets. Unlike static documentation or passive viewers, they allow users to modify code, rerun cells, and evolve analyses in real time within the ML workspace.

Why this answer

Azure Machine Learning notebooks are interactive Jupyter notebook environments hosted within Azure Machine Learning studio. They allow data scientists to write and execute Python code for data exploration, visualization, and model prototyping directly in the cloud, with built-in access to compute instances and datasets. Option B correctly identifies both the technology (Jupyter notebooks) and the primary user role (data scientists).

Exam trap

The trap here is that candidates may confuse Azure Machine Learning notebooks with generic documentation tools (Option A) or assume they are passive logs (Option C), overlooking that they are active, code-driven development environments specifically designed for data scientists.

How to eliminate wrong answers

Option A is wrong because Azure Machine Learning notebooks are not digital note-taking applications for meeting minutes; they are code-centric environments for interactive development, not documentation. Option C is wrong because notebooks are not automated logging tools that record metrics without code; logging in Azure ML requires explicit code (e.g., using `mlflow` or `run.log()`) within the notebook cells. Option D is wrong because notebooks are fully interactive read-write environments, not read-only document viewers; they allow editing and execution of code, not just review of completed results.

Practice this question →

128

MCQmedium

What is data drift in the context of deployed machine learning models?

A.When training data is accidentally deleted from storage

B.When production data distribution changes from the training data distribution over time

C.When a model's weights change during inference

D.When data is moved between different Azure storage accounts

AnswerB

Data drift occurs when the distribution of features observed in production gradually diverges from the distribution seen during training, such as when customer demographics, economic conditions, or sensor readings change over time. Because the model learned patterns from historical data, this input distribution shift causes predictions to become less accurate even though the model's logic remains unchanged. Monitoring this divergence is essential for triggering retraining.

Why this answer

Data drift refers to the phenomenon where the statistical properties of the input data a deployed model receives in production change over time, diverging from the distribution of the data used during training. This degradation can cause the model's predictions to become less accurate or unreliable, even if the model itself remains unchanged. In Azure Machine Learning, data drift is monitored using dataset monitors that compare production data distributions against the training baseline.

Exam trap

The trap here is that candidates confuse data drift with other operational issues like data loss or storage changes, rather than recognizing it as a statistical shift in the input data distribution that degrades model accuracy over time.

How to eliminate wrong answers

Option A is wrong because accidental deletion of training data is a data management or storage issue, not a change in data distribution affecting model performance. Option C is wrong because a model's weights do not change during inference; weights are fixed after training, and any change would require retraining or fine-tuning. Option D is wrong because moving data between Azure storage accounts is a data migration operation unrelated to the statistical properties of the data used for predictions.

Practice this question →

129

MCQmedium

A media company wants to automatically organize a large collection of news articles into several topic-based categories (e.g., politics, sports, technology) without using any predefined labels. They plan to use Azure Machine Learning. Which type of machine learning task should they use?

A.Regression

B.Classification

C.Clustering

D.Anomaly detection

AnswerC

Clustering is an unsupervised learning technique that partitions unlabeled data into groups based on feature similarity. By applying algorithms like k-means or hierarchical clustering, the system discovers latent topic structures in news articles without requiring predefined categories. This aligns perfectly with the scenario's goal of automatically organizing articles, as it does not rely on any labeled examples and can adapt to the inherent content patterns.

Why this answer

Clustering is the correct choice because the media company wants to group unlabeled news articles into topic-based categories based on inherent similarities in the data, without using predefined labels. Azure Machine Learning provides clustering algorithms like K-Means that automatically partition the dataset into distinct clusters, making it ideal for unsupervised learning tasks where the goal is to discover natural groupings.

Exam trap

The trap here is that candidates often confuse clustering with classification because both involve grouping data into categories, but clustering is unsupervised (no labels) while classification requires labeled training data.

Why the other options are wrong

Regression predicts a continuous numeric value, not discrete categories. The question requires grouping articles into topic categories without predefined labels, which is unsupervised learning, not regression.

Classification requires predefined labels, but the question explicitly states 'without using any predefined labels' and aims to discover topic-based categories automatically, which is a clustering task.

Anomaly detection identifies rare or unusual patterns, not grouping unlabeled data into categories. The question requires organizing articles into topic-based categories without predefined labels, which is a clustering task.

Practice this question →

130

MCQmedium

A data scientist trains a regression model on a dataset with 100 features and 10,000 samples. The model achieves a low training error but a much higher error on a held-out test set. Which approach is most likely to improve the model's generalization performance?

A.Increase the complexity of the model by adding more layers or parameters

B.Add more training data

C.Reduce the number of features or apply regularization

D.Use a different train-test split ratio like 80-20 instead of 70-30

AnswerC

Reducing the number of features or applying regularization (e.g., L1/L2) directly targets overfitting by simplifying the model and penalizing large coefficients. This forces the model to focus on generalizable patterns instead of memorizing noise in the training set, which improves test performance. With only 100 samples, regularization is especially effective because it stabilizes coefficient estimates when the feature count is relatively high.

Why this answer

The model exhibits high variance (overfitting), as indicated by low training error but high test error. Reducing the number of features or applying regularization (e.g., L1/L2) directly constrains model complexity, forcing it to learn more general patterns rather than memorizing noise. This is the standard approach to improve generalization in regression models.

Exam trap

The trap here is that candidates often assume adding more training data is always the best fix for overfitting, but the question specifically describes a model with 100 features and only 10,000 samples—feature reduction or regularization is the more direct and efficient solution.

Why the other options are wrong

The model already overfits (low training error, high test error), so increasing complexity would worsen overfitting, not improve generalization.

The model already has low training error but high test error, indicating overfitting. Adding more training data can help reduce overfitting, but the primary issue is high variance due to many features relative to samples; reducing features or regularization is more direct and effective.

The model is overfitting (low training error, high test error). Changing the train-test split ratio (e.g., from 70-30 to 80-20) does not address overfitting; it only changes the amount of data used for training and testing, which is unlikely to significantly reduce the generalization gap.

Practice this question →

131

MCQmedium

A data scientist trains a machine learning model on a dataset of housing prices. The model achieves 98% accuracy on the training data but only 72% accuracy on a separate test set. What is the most likely problem with this model?

A.Underfitting

B.Overfitting

C.Data leakage

D.Class imbalance

AnswerB

Overfitting occurs when a high-capacity model fits not only the true signal in the training data but also the random noise and idiosyncrasies unique to that sample. The model effectively memorizes training instances (or their exact decision boundaries), so training accuracy is very high; on unseen test data, those memorized patterns do not generalize, yielding much lower test accuracy. This large training-to-test gap is the classic variance-error signature, often cured by regularization, pruning, or obtaining more diverse training data.

Why this answer

The model's high accuracy on training data (98%) but significantly lower accuracy on test data (72%) is a classic symptom of overfitting, where the model learns noise and specific patterns in the training set rather than generalizing to new, unseen data. In Azure Machine Learning, this often occurs when the model is too complex (e.g., deep decision trees or high-degree polynomial features) relative to the amount of training data, and regularization techniques like L1/L2 regularization or early stopping are not applied.

Exam trap

The trap here is that candidates often confuse high training accuracy with a good model, overlooking the critical test accuracy drop that signals overfitting, and may incorrectly select underfitting because they focus only on the low test score.

Why the other options are wrong

Underfitting would cause poor performance on both training and test sets, but here training accuracy is high (98%) while test accuracy is low (72%), indicating the model memorized training data rather than failing to learn patterns.

Data leakage typically causes overly optimistic performance on both training and test sets, not a large gap between them. Here, the high training accuracy and much lower test accuracy indicate overfitting, not leakage.

Class imbalance typically causes poor performance on minority classes, not a large gap between training and test accuracy. The 98% vs 72% discrepancy indicates overfitting, not imbalance.

Practice this question →

132

MCQeasy

A data scientist trains a binary classification model to distinguish between images of cats and dogs. On the test set, the model achieves 98% accuracy, but a deeper inspection reveals that the test set contains 95% cats and 5% dogs, and the model predicts 'cat' for every single image. Which metric should the data scientist prioritize to get a more realistic evaluation of the model's performance on this imbalanced dataset?

A.Precision

B.Recall

C.F1-score

D.Accuracy

AnswerC

The F1-score combines precision and recall into a single metric that penalizes extreme values. For this model, the F1-score for the minority class (dogs) would be very low, revealing the poor performance that accuracy hides.

Why this answer

The F1-score is the harmonic mean of precision and recall, providing a single metric that balances both when classes are imbalanced. In this scenario, accuracy is misleadingly high (98%) because the model always predicts the majority class (cat), achieving high accuracy without actually learning to distinguish cats from dogs. The F1-score penalizes the model for its poor recall on the minority class (dogs), giving a more realistic evaluation of its performance.

Exam trap

The trap here is that candidates see 98% accuracy and assume the model is performing well, failing to recognize that accuracy is meaningless on imbalanced datasets where the model can achieve high accuracy by simply predicting the majority class.

Why the other options are wrong

Precision measures the proportion of positive predictions that are correct, but in this case the model predicts 'cat' for all images, so precision for 'cat' is 95% (since 95% of images are cats), which is misleadingly high and does not reveal the model's failure to detect dogs.

Recall measures the proportion of actual positives correctly identified, but here the model predicts all images as 'cat' (the majority class), so recall for 'dog' (the minority class) is 0%, which is poor. However, recall alone does not account for precision, and in this imbalanced scenario, the F1-score (harmonic mean of precision and recall) provides a more balanced evaluation.

Accuracy is misleading on imbalanced datasets because a model can achieve high accuracy by always predicting the majority class, as in this case where the model predicts 'cat' for all images and achieves 98% accuracy despite having zero predictive power for dogs.

Practice this question →

133

MCQmedium

A data scientist has a dataset containing images of handwritten digits (0-9) where each image is labeled with the correct digit. The goal is to train a model that can predict the digit from a new image. Which type of machine learning approach should be used?

A.Regression

B.Classification

C.Clustering

D.Reinforcement learning

AnswerB

Classification is a supervised learning technique that learns a decision boundary from labeled examples to assign new inputs to discrete categories. Recognizing handwritten digits is a classic multiclass classification problem where each image maps to one of ten classes. The model is trained on images with known digit labels to generalize to unseen handwriting.

Why this answer

This is a supervised learning problem where the model must predict a discrete class label (digit 0-9) from input images. Classification algorithms, such as logistic regression or neural networks, are designed to map inputs to categorical outputs, making B the correct choice.

Exam trap

The trap here is that candidates may confuse regression with classification when the output is a number (0-9), but regression is for continuous values, not discrete labels, even if the labels are numeric.

How to eliminate wrong answers

Option A is wrong because regression predicts continuous numerical values (e.g., price or temperature), not discrete categories like digits. Option C is wrong because clustering is an unsupervised learning technique that groups unlabeled data based on similarity, but here the dataset has labeled images. Option D is wrong because reinforcement learning involves an agent learning through rewards and penalties in an interactive environment, which is not applicable to static labeled image classification.

Practice this question →

134

MCQmedium

A data scientist trains a regression model to predict house prices. The model performs poorly on both the training data and the test data, showing high error in both sets. Which concept best describes this situation?

A.Overfitting

B.Underfitting

C.Data leakage

D.Feature scaling

AnswerB

Underfitting occurs when the model is too simplistic to capture the underlying relationship in the data, such as using a linear regression for a strongly nonlinear relationship or omitting key features. This high-bias state causes the model to make large, systematic errors on both the training set and the test set. The symptom of poor performance on all data splits is the hallmark of underfitting, matching the scenario described.

Why this answer

Underfitting occurs when a model is too simple to capture the underlying patterns in the data, resulting in high error on both the training and test sets. In this regression scenario, the model fails to learn the relationship between features and house prices, leading to poor performance across all data splits.

Exam trap

The trap here is that candidates confuse underfitting with overfitting because both involve poor performance, but the key distinction is that underfitting shows high error on both training and test sets, while overfitting shows low training error and high test error.

How to eliminate wrong answers

Option A is wrong because overfitting would show low error on training data and high error on test data, not high error on both. Option C is wrong because data leakage involves information from outside the training set influencing the model, which typically causes overly optimistic performance, not uniformly high error. Option D is wrong because feature scaling normalizes input ranges to improve convergence in algorithms like gradient descent, but it does not directly cause high error on both training and test sets.

Practice this question →

135

MCQmedium

What is 'Azure Machine Learning datasets' and why are they important?

A.The raw data files stored in Azure Blob Storage before any processing

B.Versioned, registered data references enabling reproducibility, sharing, and lineage tracking in Azure ML

C.Synthetic datasets automatically generated by Azure ML to supplement small training sets

D.Pre-labelled benchmark datasets provided by Microsoft for testing Azure ML models

AnswerB

In Azure ML, a dataset is a versioned, registered reference to a data source, not a copy of the data itself. It encapsulates metadata such as schema, a data profile, and sampling strategy while pointing to files in a datastore like Azure Blob Storage or Azure Data Lake. This makes experiments reproducible because each run can pin a specific dataset version, and teams can share, re-use, and trace lineage across pipelines. The decoupling from model code is what enables these governance features.

Why this answer

Azure Machine Learning datasets are versioned, registered data references that encapsulate metadata such as location, schema, and creation time, enabling reproducibility, sharing, and lineage tracking across experiments. They do not store the raw data files themselves but provide a pointer to the data source (e.g., Azure Blob Storage, Azure Data Lake), ensuring that every training run uses the exact same data snapshot, which is critical for auditability and collaboration.

Exam trap

The trap here is that candidates confuse a dataset with the raw data files themselves, assuming it is just a storage container, rather than understanding it as a versioned, registered metadata reference that enables reproducibility and lineage.

How to eliminate wrong answers

Option A is wrong because Azure ML datasets are not the raw data files themselves; they are metadata references that point to the data, and the raw files can be stored in various locations, not just Azure Blob Storage. Option C is wrong because Azure ML does not automatically generate synthetic datasets; synthetic data generation would require custom code or third-party tools, and datasets are for referencing existing data. Option D is wrong because Azure ML datasets are user-created references to their own data, not pre-labelled benchmark datasets provided by Microsoft for testing.

Practice this question →

136

MCQmedium

A data scientist is training a regression model to predict house prices in Azure Machine Learning. The model uses features like square footage, number of bedrooms, and location (zip code). The data scientist notices that the model has a very low error on the training data but a high error on the test data. Which technique should the data scientist apply during model training to reduce overfitting by penalizing large coefficients?

A.Use a smaller test set.

B.Apply feature scaling only.

C.Use a regularization algorithm like Lasso (L1).

D.Increase the number of training epochs.

AnswerC

Lasso adds an L1 penalty proportional to the absolute value of coefficients, shrinking some to zero and constraining model complexity. This penalisation reduces the overfitting causing low training error but high test error on the house price regression.

Why this answer

Lasso (L1) regularization adds a penalty equal to the absolute value of the magnitude of coefficients, which can shrink some coefficients to zero, effectively performing feature selection and reducing overfitting. This directly addresses the problem of large coefficients causing the model to fit noise in the training data, leading to high test error.

Exam trap

The trap here is that candidates often confuse regularization with feature scaling or training duration, not realizing that only regularization directly penalizes large coefficient magnitudes to combat overfitting.

Why the other options are wrong

Using a smaller test set does not address overfitting; it reduces the reliability of error estimates and may hide overfitting, but does not penalize large coefficients.

Feature scaling alone does not penalize large coefficients; it only normalizes feature ranges. Overfitting caused by large coefficients requires regularization, not scaling.

Increasing the number of training epochs does not penalize large coefficients; it can actually worsen overfitting by allowing the model to fit the training data even more closely.

Practice this question →

137

MCQmedium

What is 'stochastic gradient descent' (SGD) and how does it work?

A.A random sampling method for selecting training data without replacement

B.An optimisation algorithm that updates weights using gradients computed on random data mini-batches

C.A technique for randomly selecting which model architecture to use for AutoML

D.Randomly descending through decision tree branches to make predictions

AnswerB

Stochastic Gradient Descent (SGD) is an iterative optimization algorithm where each update step uses a randomly drawn mini-batch of training examples to compute an unbiased estimate of the loss gradient. Because the mini-batch is small and randomly chosen, the gradient estimate is noisy, but it is far cheaper to compute than the full-batch gradient, allowing SGD to scale to large datasets and still converge by taking many noisy steps.

Why this answer

Stochastic Gradient Descent (SGD) is an optimization algorithm used to train machine learning models by iteratively updating model weights. It computes the gradient of the loss function on a randomly selected mini-batch of training data (not the entire dataset), which introduces noise but significantly speeds up convergence and reduces memory usage. This mini-batch approach is the core of SGD and distinguishes it from batch gradient descent.

Exam trap

The trap here is that candidates confuse 'stochastic' with 'random sampling of data' (Option A) or 'random model selection' (Option C), when in fact SGD's stochasticity refers to using random mini-batches to compute gradients, not random data selection or architecture choice.

How to eliminate wrong answers

Option A is wrong because SGD does not sample training data without replacement; it typically samples mini-batches with replacement (or shuffles the data) to maintain stochasticity, and it is not a sampling method but an optimization algorithm. Option C is wrong because SGD is not used to select model architectures; AutoML uses techniques like Bayesian optimization, grid search, or reinforcement learning for architecture search, not gradient descent. Option D is wrong because SGD is not a decision tree traversal method; decision trees use greedy splitting criteria (e.g., Gini impurity, information gain) to make predictions, not gradient-based weight updates.

Practice this question →

138

MCQmedium

A data scientist trains a regression model to predict daily electricity consumption (in kWh) for a commercial building. The business team needs a metric that heavily penalizes large prediction errors (outliers) more than small errors. Which metric should the data scientist report to best meet this requirement?

A.Mean Absolute Error (MAE)

B.Root Mean Squared Error (RMSE)

C.R-squared

D.Mean Absolute Percentage Error (MAPE)

AnswerB

Root Mean Squared Error (RMSE) computes the square root of the mean of squared residuals, directly implementing L2 loss. Squaring each error transforms a 10-unit residual into a contribution of 100 versus 1 for a 1-unit residual, so a single large error can dominate the metric and drive the model to reduce it. This quadratic amplification is exactly the heavier penalty on outliers required, making RMSE the correct metric when large errors are especially undesirable.

Why this answer

Root Mean Squared Error (RMSE) is the correct metric because it squares the residuals before averaging, which disproportionately amplifies the impact of large errors (outliers) compared to small errors. This aligns directly with the business requirement to heavily penalize large prediction errors in the regression model for daily electricity consumption.

Exam trap

The trap here is that candidates often confuse MAE as a robust metric for all error scenarios, but the question explicitly requires heavy penalization of outliers, which only RMSE (or MSE) achieves through squaring errors.

How to eliminate wrong answers

Option A is wrong because Mean Absolute Error (MAE) treats all errors linearly, giving equal weight to small and large errors, so it does not heavily penalize outliers. Option C is wrong because R-squared measures the proportion of variance explained by the model, not the magnitude or penalty of prediction errors, and it does not specifically penalize outliers. Option D is wrong because Mean Absolute Percentage Error (MAPE) uses percentage-based errors, which can be unstable when actual values are near zero and does not inherently square or amplify large errors more than small ones.

Practice this question →

139

MCQmedium

A data scientist is training a classification model on a dataset with 100 features and only 500 labeled samples. The model achieves 99% accuracy on the training data but only 68% accuracy on a held-out test set, indicating overfitting. Which technique is most appropriate to directly address this problem?

A.Increase the amount of training data by collecting more samples

B.Reduce the number of features used for training

C.Increase the complexity of the model by adding more layers

D.Train for more epochs

AnswerB

Reducing the number of features (e.g., via feature selection or principal component analysis) directly lowers the dimensionality of the input space, which reduces the model's capacity to fit idiosyncratic noise in the training data. With fewer parameters to estimate, the model learns simpler, more generalizable decision boundaries instead of memorizing spurious correlations. This is a standard regularization technique, particularly effective when the number of features exceeds the number of samples, and it often improves test accuracy even if training accuracy drops slightly.

Why this answer

Reducing the number of features directly combats overfitting by decreasing model complexity and the risk of learning noise from irrelevant or redundant features. With only 500 samples and 100 features, the model has a high variance problem; feature selection or dimensionality reduction (e.g., using Azure Machine Learning's Filter-Based Feature Selection or PCA) simplifies the hypothesis space, improving generalization to the test set.

Exam trap

The trap here is that candidates may assume more data (Option A) is always the best fix for overfitting, but the question explicitly tests the ability to choose a technique that directly addresses the high-dimensional, low-sample scenario without requiring additional data collection.

Why the other options are wrong

The model already overfits with 500 samples; adding more data would help generalization but is not the most direct technique to address overfitting caused by high dimensionality relative to sample size.

Increasing model complexity by adding more layers would exacerbate overfitting, especially given the small sample size relative to the number of features, making the model even more prone to memorizing noise.

Training for more epochs would likely increase overfitting further, as the model already memorizes the training data (99% accuracy vs 68% test accuracy). More epochs exacerbate overfitting rather than reduce it.

Practice this question →

140

MCQmedium

What is 'label imbalance' in a classification dataset and how does it affect model training?

A.When labels in the training data contain spelling errors

B.When one class greatly outnumbers others, causing models to be biased toward the majority class

C.When training labels are applied inconsistently by different human annotators

D.When a model produces predictions that don't match any of the training labels

AnswerB

Class imbalance occurs when one class, such as 'non-fraud,' vastly outnumbers another, such as 'fraud.' A model trained on such data can trivially achieve high accuracy by always predicting the majority class, while failing to recognize the minority class that is often the most important to detect. Mitigations include resampling methods like oversampling the minority class or undersampling the majority class, applying class weights to the loss function, and evaluating with precision, recall, or F1-score instead of raw accuracy.

Why this answer

Label imbalance refers to a situation in classification datasets where one class (the majority class) has significantly more samples than other classes (minority classes). This causes the model to become biased toward predicting the majority class, as it minimizes overall loss by ignoring minority classes, leading to poor generalization and low recall for underrepresented classes.

Exam trap

The trap here is that candidates confuse label imbalance with data quality issues like label noise or annotation errors, leading them to pick options A or C instead of recognizing it as a class distribution problem.

How to eliminate wrong answers

Option A is wrong because spelling errors in labels are a data quality issue, not a class distribution imbalance; they relate to data cleaning, not the relative frequency of classes. Option C is wrong because inconsistent labeling by annotators is an inter-annotator agreement problem, which affects label noise and reliability, not the proportional count of samples per class. Option D is wrong because predictions that don't match training labels describe a model's inability to map to known classes (e.g., out-of-distribution detection), not an imbalance in the training data's class distribution.

Practice this question →

141

MCQhard

A data scientist is using Azure Automated Machine Learning to build a binary classification model for a highly imbalanced dataset (95% negative, 5% positive). The data scientist wants AutoML to select the best model based on a metric that is robust to class imbalance. Which primary metric should the data scientist configure in the AutoML settings?

A.Accuracy

B.AUC_weighted

C.F1_score

D.Log_loss

AnswerB

AUC_weighted computes the area under the ROC curve separately for each class and then aggregates these values using the proportion of true samples in each class as weights. Because the ROC curve evaluates model ranking across all probability thresholds without fixation on a single decision boundary, AUC_weighted remains robust to severe class imbalance and is the AutoML default primary metric for imbalanced classification problems. This weighting ensures the minority class's separability contributes proportionally to the score, so model selection is not dominated by majority-class accuracy.

Why this answer

AUC_weighted is the correct primary metric for imbalanced binary classification because it computes the area under the ROC curve for each class and averages them with weight proportional to the class support. This weighting ensures that the metric reflects performance on both the majority (95% negative) and minority (5% positive) classes, making it robust to severe imbalance. Azure Automated Machine Learning uses AUC_weighted as a recommended metric when the dataset is skewed, as it penalizes models that ignore the minority class.

Exam trap

The trap here is that candidates often choose Accuracy because it is the most intuitive metric, failing to recognize that on imbalanced datasets it can be misleadingly high and does not reflect minority class performance.

Why the other options are wrong

Accuracy is not robust to class imbalance because a model that predicts the majority class for all instances can achieve 95% accuracy, failing to identify any positive cases.

Log loss measures the average log-likelihood of predictions, but it is not robust to class imbalance because it treats all errors equally, and the majority class dominates the loss, making it insensitive to improvements in minority class recall.

Practice this question →

142

MCQmedium

What is model monitoring in Azure Machine Learning and why is it important?

A.Checking how many API calls the model endpoint receives per hour

B.Tracking model prediction quality and data distribution changes in production to detect degradation

C.Monitoring the GPU memory usage during model training

D.Reviewing model architecture choices for optimization

AnswerB

This is the core of model monitoring: systematically observing a deployed model's prediction performance and the statistical nature of its inputs over time to detect degradation. It involves tracking quality metrics against ground truth when available and running distributional comparisons (such as PSI or KS tests) to identify data drift and concept drift. When degradation is detected, the model can be flagged for retraining or investigation, ensuring continued reliability in production.

Why this answer

Model monitoring in Azure Machine Learning is the continuous tracking of a deployed model's performance in production, focusing on prediction quality (e.g., accuracy, precision, recall) and data distribution shifts (data drift) to detect degradation over time. This is critical because models can become stale as real-world data evolves, leading to poor business decisions or compliance failures. Azure ML's Model Data Collector and monitoring dashboards automatically capture input data and predictions, alerting data scientists when drift or performance drops below defined thresholds.

Exam trap

The trap here is that candidates confuse operational metrics (like API call count or GPU usage) with model-specific performance monitoring, leading them to pick options that describe infrastructure monitoring rather than model quality tracking.

How to eliminate wrong answers

Option A is wrong because checking API call volume is a metric for endpoint usage or load, not model monitoring; it does not assess prediction quality or data drift. Option B is correct as described. Option C is wrong because monitoring GPU memory during training is part of training infrastructure optimization, not production model monitoring.

Option D is wrong because reviewing model architecture is a design-time activity, not a post-deployment monitoring task.

Practice this question →

143

MCQmedium

What is 'model monitoring' in Azure Machine Learning after deployment?

A.Watching the training loss curve during model training to detect overfitting

B.Tracking deployed model performance and data drift over time to detect degradation

C.A dashboard showing the compute costs of running model inference in production

D.Monitoring the uptime and latency of the model serving endpoint

AnswerB

Tracking deployed model performance and data drift over time is the essence of model monitoring: it continuously compares live inference data against the training dataset to detect feature-distribution drift and evaluates prediction quality using metrics like accuracy or AUC when ground truth is available. When drift triggers a threshold, it alerts data scientists to a silent degradation that could warrant retraining. This is a post-deployment, production-focused practice that Azure Machine Learning supports natively.

Why this answer

Model monitoring in Azure Machine Learning refers to the ongoing process of tracking a deployed model's performance metrics (such as accuracy or precision) and detecting data drift (changes in input data distribution) or concept drift (changes in the relationship between inputs and outputs) over time. This is critical because models can degrade in production even if they performed well during training, due to shifts in real-world data. Azure ML provides built-in monitoring capabilities, including drift detection and alerting, to ensure models remain reliable.

Exam trap

The trap here is that candidates confuse infrastructure monitoring (uptime/latency) or cost tracking with model-specific monitoring (performance and drift), which is the core focus of 'model monitoring' in Azure ML.

How to eliminate wrong answers

Option A is wrong because watching the training loss curve during model training is part of training diagnostics, not post-deployment monitoring; it detects overfitting during training, not production degradation. Option C is wrong because a dashboard showing compute costs is a cost management feature, not model monitoring; it tracks resource usage, not model performance or data drift. Option D is wrong because monitoring endpoint uptime and latency is infrastructure monitoring (DevOps/MLOps concern), not model monitoring; it ensures availability but does not detect performance degradation or drift in the model's predictions.

Practice this question →

144

MCQhard

What is 'neural architecture search' (NAS) and how does it relate to AutoML?

A.Searching the web for neural network architectures published in research papers

B.Automating the discovery of optimal neural network architectures using computational search

C.Querying a database of pre-built neural networks to find the closest match for a task

D.A legal search process for patenting new AI model architectures

AnswerB

Neural Architecture Search (NAS) is an automated, iterative optimization process that explores a defined space of possible network designs—such as layer types, depth, width, and connectivity—using methods like reinforcement learning, evolutionary algorithms, or gradient-based approaches. Each candidate architecture is trained and evaluated on validation data, and the search algorithm uses those performance signals to propose better candidates. This allows NAS to discover novel, high-performing architectures that may surpass human-designed networks, making it a computational search rather than a retrieval or manual process.

Why this answer

Neural Architecture Search (NAS) is an automated process that uses computational search methods—such as reinforcement learning, evolutionary algorithms, or gradient-based optimization—to discover optimal neural network architectures for a given task. It is a key component of AutoML because AutoML aims to automate the entire machine learning pipeline, including model selection and hyperparameter tuning, and NAS specifically automates the design of the neural network topology itself.

Exam trap

The trap here is that candidates confuse NAS with simply searching for existing models online or in a database, rather than understanding it as an automated, generative search process that creates new architectures.

How to eliminate wrong answers

Option A is wrong because NAS does not involve searching the web for published research papers; it is a computational search over a defined architecture space, not a web crawl. Option C is wrong because NAS does not query a static database of pre-built networks; it dynamically generates and evaluates candidate architectures during the search process. Option D is wrong because NAS is a technical optimization method, not a legal or patent-related search process.

Practice this question →

145

MCQmedium

A bike-sharing company wants to predict the number of rentals per hour. Their model's predictions are usually close but occasionally have large errors due to unexpected events like sudden rain. They want a metric that heavily penalizes these large errors to ensure the model is not overly confident. Which evaluation metric should they primarily use?

A.Mean Absolute Error (MAE)

B.Mean Squared Error (MSE)

C.Classification Accuracy

D.R-squared

AnswerB

Mean Squared Error squares each residual before averaging, so a prediction that is off by 10 contributes 100 units to the loss while an error of 2 contributes only 4. This quadratic weighting is exactly what the company wants: large errors are penalized disproportionately, preventing a model that tolerates occasional severe under- or over-prediction. Because MSE is also differentiable, it is a natural loss function for gradient-descent-based regression training.

Why this answer

Mean Squared Error (MSE) is the correct choice because it squares the residuals, which heavily penalizes large errors. Since the bike-sharing company wants to discourage occasional large prediction errors (e.g., due to sudden rain), MSE’s quadratic penalty ensures that models with even a few large outliers receive a much worse score, forcing the model to avoid overconfidence.

Exam trap

The trap here is that candidates often choose MAE because it is simpler and more interpretable, but they miss the explicit requirement to 'heavily penalize large errors,' which only MSE (or RMSE) accomplishes through squaring.

Why the other options are wrong

MAE does not heavily penalize large errors because it treats all errors linearly, whereas the question requires a metric that penalizes large errors more, which is a key property of MSE.

Classification Accuracy is used for classification tasks, not regression. The question asks about predicting a continuous number (rentals per hour), so accuracy is inappropriate.

R-squared measures the proportion of variance explained by the model, not the penalty for large errors. It does not heavily penalize occasional large errors like MSE does.

Practice this question →

146

MCQeasy

What is 'Azure ML's experiment tracking' and why do data scientists use it?

A.Monitoring the progress of Azure ML service new feature deployments

B.Recording hyperparameters, metrics, and configurations for each training run for comparison and reproduction

C.Tracking which Azure ML resources are used by which team members for billing allocation

D.A compliance audit log of all model predictions made in production

AnswerB

Recording hyperparameters, metrics, and configurations for each training run is exactly what Azure ML experiment tracking does. Each run—often managed through Azure ML's SDK, CLI, or MLflow integration—logs a unique snapshot of the code, environment, dataset version, and hyperparameters, along with scalar and tensor metrics like loss or accuracy. This enables data scientists to compare runs side-by-side in the studio, identify the best performing configuration, and reproduce historical results by replaying the exact logged run. The Run History and metrics tables are purpose-built for this iterative, experimental comparison during model development.

Why this answer

Azure ML's experiment tracking is a feature that automatically records hyperparameters, metrics, and configuration details for each training run. Data scientists use it to compare multiple runs, identify the best-performing model, and reproduce results by revisiting the exact settings and data used. This is essential for iterative experimentation and ensuring reproducibility in machine learning workflows.

Exam trap

The trap here is that candidates confuse experiment tracking (recording training run metadata) with monitoring or auditing of deployed models, leading them to choose options about deployment progress or production compliance logs.

How to eliminate wrong answers

Option A is wrong because it describes monitoring deployment progress of new features, which is a DevOps or MLOps concern, not the purpose of experiment tracking for training runs. Option C is wrong because it refers to resource usage tracking for billing allocation, which is handled by Azure Cost Management and resource tagging, not by experiment tracking. Option D is wrong because it describes a compliance audit log for model predictions in production, which is related to model monitoring and governance, not the recording of training run metadata.

Practice this question →

147

MCQeasy

What is machine learning?

A.A process of manually programming computers with rules for every possible scenario

B.A subset of AI where algorithms learn from data to make predictions without explicit programming

C.A method of creating robots that can perform physical tasks

D.A type of computer network for processing large datasets

AnswerB

Machine learning is a subset of artificial intelligence in which algorithms iteratively adjust internal parameters by minimizing a loss function over training data, thereby learning patterns that generalize to new, unseen inputs. This is fundamentally statistical rather than rule-based: the algorithm infers the mapping from features to outputs automatically, without human-provided step-by-step instructions. Core paradigms include supervised learning, unsupervised learning, and reinforcement learning, each relying on data-driven optimization rather than explicit programming.

Why this answer

Machine learning is a subset of artificial intelligence (AI) that enables systems to automatically learn and improve from experience without being explicitly programmed for every scenario. Instead of following static rules, ML algorithms use training data to identify patterns and make predictions or decisions. This is the core definition tested in AI-900, distinguishing ML from traditional rule-based programming.

Exam trap

The trap here is that candidates confuse machine learning with traditional programming (Option A) because both involve computers making decisions, but ML eliminates the need for explicit rule-writing by learning from data.

How to eliminate wrong answers

Option A is wrong because it describes traditional rule-based programming, not machine learning; ML does not require manual coding of rules for every possible scenario but instead learns patterns from data. Option C is wrong because machine learning is not limited to robotics or physical tasks; it is a data-driven approach used in software applications like recommendation systems and fraud detection. Option D is wrong because while machine learning may use computer networks for processing large datasets, this describes distributed computing or big data infrastructure, not the fundamental concept of learning from data to make predictions.

Practice this question →

148

MCQmedium

A manufacturer trains a model to detect defective parts on an assembly line. Only 2% of parts are defective. The model predicts 'non-defective' for all parts and achieves 98% accuracy. Which metric best reveals the model's inability to identify defective parts?

A.Accuracy

B.Precision

C.Recall

D.F1 Score

AnswerC

Recall, also called sensitivity or the true positive rate, is computed as TP / (TP + FN), where FN represents actual defective parts that the model failed to identify. Because the model detected no defects, TP = 0, so recall is 0%, which correctly reveals that the model misses all defective parts. This makes recall the most direct and appropriate metric for a defect-detection problem where the cost of missing a defect is high.

Why this answer

Recall (sensitivity) measures the proportion of actual defective parts correctly identified by the model. With 98% accuracy but zero true positives (since the model labels everything as non-defective), recall is 0%, which directly exposes the model's failure to detect any defective parts despite high accuracy.

Exam trap

The trap here is that candidates see 98% accuracy and assume the model is performing well, overlooking that accuracy is inflated by class imbalance and does not measure the model's ability to detect the rare defective class.

How to eliminate wrong answers

Option A is wrong because accuracy is misleading here—it only reflects the overall correct predictions (98% non-defective) and hides the model's complete failure on the minority class (defective parts). Option B is wrong because precision measures the proportion of predicted defective parts that are actually defective; since the model never predicts defective, precision is undefined (division by zero) and does not reveal the inability to identify defects. Option D is wrong because the F1 score is the harmonic mean of precision and recall; with recall at 0%, F1 is also 0%, but recall alone more directly and intuitively shows the model's inability to detect defects.

Practice this question →

149

MCQhard

What is 'curriculum learning' and how does it relate to training stability?

A.Designing a course curriculum using AI to personalise learning for students

B.Training models on progressively harder examples to improve stability and convergence

C.A structured plan for the sequence of ML courses a data scientist should take

D.Using a pre-defined curriculum of hyperparameter values to systematically explore the search space

AnswerB

Curriculum learning is a training strategy that orders examples from easy to hard rather than presenting them randomly. By starting with simple, low-noise samples, the model's early gradients are more stable, which helps it avoid sharp local minima and makes the optimization landscape easier to traverse. This progressive difficulty acts as a form of implicit regularization, often yielding faster convergence and better generalization on the final task.

Why this answer

Curriculum learning is a training strategy where a model is first exposed to simpler examples and then gradually introduced to more complex ones. This approach improves training stability by preventing the model from being overwhelmed by difficult patterns early on, which can cause large gradient updates and divergence. By structuring the learning process, the model converges more reliably and often achieves better generalization.

Exam trap

The trap here is that candidates confuse 'curriculum learning' with educational curricula or hyperparameter tuning, because the term 'curriculum' sounds like a course plan or a search schedule rather than a data ordering strategy.

How to eliminate wrong answers

Option A is wrong because it describes adaptive educational technology for human learners, not a machine learning training technique. Option C is wrong because it refers to a sequence of courses for a data scientist's professional development, not a model training methodology. Option D is wrong because it describes a hyperparameter search strategy (like grid or random search), not a curriculum-based ordering of training examples.

Practice this question →

150

MCQmedium

What is 'data augmentation' and how does it help with limited training data?

A.Collecting more labelled data from external sources to supplement training

B.Creating synthetic training variants (flips, rotations, synonyms) to expand small datasets

C.Increasing the number of compute nodes to process large training datasets faster

D.Adding more evaluation metrics to get a richer view of model performance

AnswerB

Creating synthetic training variants — such as image flips/rotations or text synonym substitution — is the core of data augmentation. These transformations generate new, plausible examples from existing ones, multiplying effective dataset size while teaching the model invariances (e.g., a cat is still a cat after horizontal flip) and reducing overfitting. This is exactly the technique used to expand small datasets in computer vision and NLP.

Why this answer

Data augmentation is a technique that artificially expands a training dataset by applying transformations (e.g., image flips, rotations, cropping, or text synonym replacement) to existing samples. This helps models generalize better when real-world data is scarce, reducing overfitting without requiring new labeled data collection.

Exam trap

The trap here is that candidates confuse 'data augmentation' with simply 'collecting more data' (Option A), failing to recognize that augmentation creates synthetic variants from existing data rather than acquiring new external samples.

How to eliminate wrong answers

Option A is wrong because collecting more labeled data from external sources is a separate process (data acquisition), not data augmentation—augmentation creates synthetic variants from existing data, not new external samples. Option C is wrong because increasing compute nodes relates to distributed training or scaling infrastructure, not to generating synthetic training variants to address limited data. Option D is wrong because adding evaluation metrics (e.g., precision, recall) improves model assessment but does not expand the training dataset or solve data scarcity.

Practice this question →