AI0-001 AI Models and Data Engineering Practice Test 4 — 15 Questions

Question 1

A data engineer is preparing a dataset for training a classification model. The dataset contains missing values in multiple features, inconsistent categorical labels, and outliers in numerical features. Which TWO preprocessing steps should the engineer prioritize to improve model performance?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Normalize numerical features using min-max scaling.. Option A is correct because min-max scaling normalizes numerical features to a fixed range (typically [0, 1]), which reduces the disproportionate influence of outliers and features with larger magnitudes, helping distance-based and gradient-based classifiers converge better. Option D is correct because imputing missing values with the median is a robust strategy for numerical features: unlike the mean, the median is resistant to outliers, so it preserves the central tendency of the data while avoiding the data loss and bias that come from dropping rows. Option B is not appropriate because removing every row with any missing value can discard a large portion of the dataset and introduce selection bias, which is usually worse than imputation. Option C is not the best choice here because label encoding imposes an arbitrary ordinal relationship on nominal categories, which can mislead classifiers; it is only suitable for genuinely ordinal variables. Option E is also not universally correct because one-hot encoding all categorical variables can cause high dimensionality and sparsity, and it is unnecessary for ordinal features or low-cardinality variables where other encodings suffice.

Answer

Remove all rows with any missing data.

Answer

Encode categorical variables using label encoding.

Answer

Apply one-hot encoding to all categorical variables.

Question 2

Refer to the exhibit. A data scientist reviews the pipeline and notes that the model performance degraded. Which change to the pipeline would most likely improve model performance?

Accepted Answer

Change the impute strategy from mean to median for the 'income' column.. The 'income' column likely contains outliers that skew the mean, making mean imputation sensitive to extreme values. Using median imputation is more robust to outliers, preserving the central tendency of the data and reducing bias introduced during preprocessing. This directly addresses the model performance degradation caused by distorted feature distributions.

Answer

Remove the normalization step entirely.

Answer

Drop the 'product_category' column instead of one-hot encoding.

Answer

Change the encoding method from onehot to label encoding.

Question 3

A retail company uses a machine learning model to predict daily sales. The model takes features like past sales, promotions, holidays, and weather data. Recently, the model's accuracy dropped significantly. The data engineer checks the data pipeline and finds that the weather data source changed from a free API to a new paid API that provides more detailed data. The new data includes additional attributes like humidity and wind speed, but the existing pipeline only ingests temperature and precipitation. Also, the time zone format changed from UTC to local time. The model was trained on the old format. Which action should the engineer take first to restore model performance?

Accepted Answer

Transform the new data to match the old format (time zone and selected features) and retrain the model.. The model's performance dropped because the input data distribution changed: the time zone format shifted from UTC to local time and the feature set changed. The fastest way to restore performance is to transform the new data back to the exact format the model was trained on (UTC, temperature and precipitation only) and then retrain the model on the corrected pipeline. This addresses the root cause — data drift — without introducing new variables.

Answer

Add a new step to merge old and new weather data before feeding to the model.

Answer

Revert to the old weather API.

Answer

Retrain the model with the new data including all new features.

Question 4

A data engineer is designing a pipeline to ingest transactional data from multiple sources into a data lake. The data arrives in various formats (JSON, CSV, Parquet) and requires schema enforcement and versioning. Which storage format is most suitable for the data lake to support efficient querying and schema evolution?

Accepted Answer

Parquet. Parquet is a columnar format that offers high compression and efficient querying by reading only needed columns. It supports schema evolution by allowing new columns to be added, and when used with modern table formats, it provides versioning and ACID transactions. This makes it the optimal choice for a data lake requiring schema enforcement, versioning, and efficient querying across multiple formats.

Answer

Avro

Answer

JSON

Answer

CSV

Question 5

A machine learning engineer is building a model to detect fraudulent credit card transactions. The dataset has 1,000,000 transactions, of which only 0.2% are fraudulent. The team wants to ensure the model can effectively identify fraudulent cases without being overwhelmed by the majority class. Which technique is most appropriate to address this class imbalance during training?

Accepted Answer

Apply SMOTE (Synthetic Minority Over-sampling Technique) to generate synthetic fraudulent transactions.. SMOTE is the most appropriate because it synthetically generates new minority class examples, directly addressing the scarcity of fraudulent transactions. This allows the model to learn a more generalizable decision boundary for the minority class without discarding majority class data, which is crucial when the imbalance is severe.

Answer

Use a weighted loss function that assigns a higher misclassification cost to the minority class.

Answer

Randomly undersample the majority class to match the minority class size.

Answer

Collect more data to increase the number of fraudulent transactions.

Question 6

A data engineer is designing a feature store for a fraud detection model that requires both real-time transaction features (computed from the last 5 minutes of activity) and historical batch features (computed daily). The team needs consistent feature values between training and inference to avoid training-serving skew. Which component of the feature store architecture is MOST critical to ensure consistency?

Accepted Answer

A point-in-time correct offline store that stores historical feature values and serves them for training with timestamps.. The critical component is a point-in-time correct offline store. When training a fraud model, features like 'number of transactions in last 5 minutes' must be computed as of the transaction time, not using future data. The offline store must serve historical feature values with timestamps to match the online store's real-time behavior. This prevents leakage and ensures the model sees the same feature distributions during training and inference.

Answer

A low-latency online store that caches the most recent feature values for real-time inference.

Answer

A feature transformation pipeline that applies the same code to both batch and streaming data.

Answer

A metadata registry that tracks feature definitions, owners, and versions.

Question 7

A machine learning engineer is preparing a dataset for a model that predicts whether a customer will make a purchase within the next week. The dataset contains 100,000 records with features such as age, income, and browsing history. The target variable is highly imbalanced: only 1% of customers make a purchase. The engineer plans to address the imbalance. Which TWO techniques are appropriate for handling this class imbalance? (Choose two.)

Accepted Answer

Use a weighted loss function that assigns a higher misclassification cost to the minority class.. Class imbalance can be addressed at the data level (e.g., SMOTE) or at the algorithm level (e.g., weighted loss). SMOTE generates synthetic minority samples to balance the dataset, while weighted loss penalizes misclassification of the minority class more heavily. Both are effective and widely used. Removing majority samples or duplicating minority samples are less effective and can harm model performance.

Answer

Apply principal component analysis (PCA) to reduce dimensionality and then train the model.

Answer

Increase the size of the dataset by duplicating minority class samples.

Answer

Remove all majority class samples until the classes are balanced.

Question 8

A data engineer is designing a pipeline to handle missing values in a large dataset used for training a fraud detection model. The dataset has numerical features with about 15% missing values that are missing completely at random (MCAR). The engineer wants to avoid introducing bias and preserve the distribution of the data. Which imputation method is most appropriate?

Accepted Answer

Multiple imputation by chained equations (MICE). Multiple imputation by chained equations (MICE) is the most appropriate method because it models each feature with missing values based on other features, preserving relationships and reducing bias. It generates multiple imputed datasets to account for uncertainty. Mean imputation distorts distributions, listwise deletion loses data, and KNN imputation can be computationally intensive and less effective for large datasets. Therefore, MICE is the best choice for handling MCAR data in this fraud detection scenario.

Answer

Listwise deletion

Answer

Mean imputation

Answer

K-nearest neighbors (KNN) imputation

Question 9

A machine learning team is deploying a model that predicts the remaining useful life (RUL) of industrial equipment. The model is trained on sensor data collected over time. The team wants to ensure the model generalizes well to new equipment units not seen during training. Which validation strategy is most appropriate?

Accepted Answer

Group k-fold cross-validation, where each group corresponds to a unique equipment unit.. When data contains groups (e.g., multiple sensor readings per equipment unit), it is crucial to keep all data from a group in the same fold to avoid leakage. Group k-fold cross-validation does exactly that, providing an unbiased estimate of model performance on unseen groups. This is essential for deployment on new equipment units.

Answer

Time series split cross-validation, where folds are split chronologically.

Answer

Standard k-fold cross-validation, where data is randomly split into k folds.

Answer

Leave-one-out cross-validation, where each sample is used once as validation.

Question 10

A data scientist is preparing a dataset for a machine learning model to predict customer satisfaction scores. The dataset contains several numerical features, including 'age', 'income', and 'purchase_frequency', as well as a categorical feature 'preferred_channel' with values like 'email', 'phone', and 'in-store'. The team needs to preprocess these features appropriately before training a linear regression model. Which TWO actions should the data scientist take? (Choose two.)

Accepted Answer

Apply one-hot encoding to the 'preferred_channel' feature to convert it into numerical binary columns.. One-hot encoding is essential to convert the nominal 'preferred_channel' into a usable numerical format for linear regression. Standardizing the numerical features ensures they are on comparable scales, which benefits optimization and regularization. Together, these two actions properly prepare the dataset for training a linear regression model.

Answer

Remove the 'age' feature because it is highly correlated with 'income'.

Answer

Apply a logarithmic transformation to 'purchase_frequency' to normalize its distribution.

Answer

Replace missing values in 'income' with the mean of the entire column.

Question 11

A data scientist is evaluating a binary classification model on a test set with 1,000 samples, where 950 are negative and 50 are positive. The model predicts 40 true positives, 10 false negatives, 100 false positives, and 850 true negatives. The scientist wants to choose a metric that best reflects the model's ability to correctly identify positive cases while minimizing false alarms. Which metric should the scientist use?

Accepted Answer

F1 score. The F1 score balances precision and recall, making it ideal for imbalanced datasets where both false positives and false negatives matter. Accuracy is misleading due to class imbalance, precision ignores false negatives, and recall ignores false positives. The scientist wants to identify positive cases while minimizing false alarms, so F1 score is the best choice. It provides a single metric that captures the trade-off between precision and recall.

Answer

Recall

Answer

Precision

Answer

Accuracy

Question 12

An engineer is building a regression model to predict housing prices. The dataset includes features such as square footage, number of bedrooms, and year built. The engineer notices that the square footage values range from 500 to 10,000, while the number of bedrooms ranges from 1 to 5. Which preprocessing step is most critical before training a gradient descent-based model?

Accepted Answer

Normalize or standardize the features. Gradient descent-based models are sensitive to the scale of input features because they update weights proportionally to the gradient, which is influenced by feature magnitudes. With square footage ranging 500–10,000 and bedrooms 1–5, the larger feature will dominate the gradient, causing slow or unstable convergence. Normalizing or standardizing (e.g., Z-score or min-max scaling) ensures all features contribute equally, leading to faster and more reliable training.

Answer

Use k-fold cross-validation

Answer

Apply log transformation to all features

Answer

One-hot encode the features

Question 13

A machine learning team is deploying a sentiment analysis model for customer reviews. The model was trained on reviews from an e-commerce site but will be used for a social media platform. The team observes a drop in accuracy. Which concept best explains this issue?

Accepted Answer

Data drift. Data drift occurs when the statistical properties of the input data change between the training and production environments. Here, the model was trained on e-commerce reviews but is now processing social media posts, which have different vocabulary, tone, and structure, causing a mismatch in the input distribution and leading to accuracy degradation.

Answer

Concept drift

Answer

Bias-variance tradeoff

Answer

Overfitting

Question 14

A data engineer needs to design a data pipeline for a real-time fraud detection system. The system requires low-latency processing of streaming transactions. Which architecture is most appropriate?

Accepted Answer

Stream processing with Apache Kafka and Flink. Apache Kafka provides a distributed, fault-tolerant event streaming platform that ingests high-throughput transaction data with low latency, while Apache Flink offers true stream processing with exactly-once semantics and sub-second event-time processing. Together, they enable real-time fraud detection by analyzing transactions as they arrive, without the delays inherent in batch or micro-batch approaches.

Answer

Data lake with Apache Spark

Answer

Batch processing with Apache Hadoop

Answer

Microservices architecture with REST APIs

Question 15

A team is training a deep learning model for image classification. The training loss decreases rapidly but validation loss starts increasing after a few epochs. Which regularization technique should be applied to mitigate this issue?

Accepted Answer

Early stopping. Early stopping halts training when validation loss starts increasing, preventing overfitting. Option A (data augmentation) is wrong because it increases data diversity but does not stop training when validation loss increases. Option B (L2 regularization) is wrong because it penalizes large weights but does not directly address the issue of validation loss increasing. Option D (dropout) is wrong because while it helps generalize by randomly dropping neurons, it does not stop training when overfitting occurs.

Answer

Data augmentation

Answer

L2 regularization

Answer

Dropout