AI0-001 AI Models and Data Engineering Practice Test 1 — 15 Questions

Question 1

A data scientist is preparing a dataset for training a classification model. The dataset contains 10,000 records with a binary target variable where 9,500 belong to class A and 500 belong to class B. Which technique should the scientist use to address the class imbalance?

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

Accepted Answer

SMOTE (Synthetic Minority Oversampling Technique). SMOTE is the correct technique because it generates synthetic samples for the minority class (class B) by interpolating between existing minority instances, effectively balancing the dataset without losing information. This approach avoids the overfitting risk of simple oversampling and the information loss of undersampling, making it ideal for a 19:1 imbalance ratio.

Answer

Random undersampling of class A

Answer

Adding Gaussian noise to class B

Answer

Principal Component Analysis (PCA)

Question 2

An organization needs to store sensitive customer data for training a machine learning model. The data must be encrypted at rest and in transit, and access must be audited. Which combination of practices should be implemented?

Accepted Answer

Use TLS for transfer, AES-256 for storage, and AWS CloudTrail for auditing. It combines TLS (Transport Layer Security) for encrypting data in transit, AES-256 for strong encryption at rest, and AWS CloudTrail for auditing API-level access. TLS ensures confidentiality and integrity during transmission, AES-256 provides robust symmetric encryption for stored data, and CloudTrail logs all AWS API calls for compliance and audit trails. This triad satisfies the requirements of encryption in transit, at rest, and audited access.

Answer

Use FTP for transfer, AES-128 for storage, and manual log review

Answer

Use SSH for transfer, store data in a database, and enable access logs

Answer

Use MD5 for hashing, store data in plaintext, and enable server logs

Question 3

An engineer is training a neural network and observes the output shown. Which conclusion is most likely correct?

Accepted Answer

The model is overfitting after epoch 2.. The output shows training loss decreasing while validation loss increases after epoch 2, which is a classic sign of overfitting. The model begins to memorize the training data rather than generalize, leading to poor performance on unseen data. This pattern confirms that overfitting starts after epoch 2, making option B correct.

Answer

The gradients are vanishing.

Answer

The model is underfitting.

Answer

The learning rate is too high.

Question 4

A data engineer is reviewing an S3 bucket policy for a machine learning project. The policy is intended to allow access to training data only from the corporate network (10.0.0.0/16). However, users in the corporate network report access denied. Which issue is most likely causing the problem?

Accepted Answer

The policy does not include s3:ListBucket action, which is needed to list objects.. The [CORRECT] answer is C because S3 bucket policies require both object-level actions (like s3:GetObject) and bucket-level actions (like s3:ListBucket) to be explicitly allowed. When users attempt to access objects via the console or list contents, the absence of s3:ListBucket results in an access denied error, even if s3:GetObject is allowed. This is a common oversight in S3 policies where the focus is on object permissions but the bucket-level list permission is forgotten.

Answer

The policy is missing a Deny statement.

Answer

The resource ARN is incorrect.

Answer

The IP address condition is not matching because the corporate network uses a different CIDR.

Question 5

A data scientist is training a deep learning model for image classification. The training loss decreases steadily but the validation loss starts increasing after 10 epochs. Which technique should the scientist apply to address this issue?

Accepted Answer

Implement early stopping. The scenario describes overfitting: the model memorizes training data (loss decreases) but fails to generalize to unseen validation data (validation loss increases). Early stopping (Option C) halts training when validation performance degrades, preventing overfitting while preserving the best model weights. This is a standard regularization technique in deep learning frameworks like TensorFlow and PyTorch.

Answer

Add more dropout layers

Answer

Reduce the learning rate

Answer

Increase the number of training epochs

Question 6

A financial institution is building a fraud detection system using a supervised learning model. The dataset is highly imbalanced with 99.9% legitimate transactions and 0.1% fraudulent ones. Which approach would be MOST effective to train the model to detect fraud?

Accepted Answer

Use SMOTE to generate synthetic fraudulent transactions. SMOTE (Synthetic Minority Oversampling Technique) is the most effective approach because it generates synthetic fraudulent transactions by interpolating between existing minority class samples, thereby balancing the dataset without losing information. This allows the model to learn decision boundaries for fraud detection more effectively than simple undersampling or metric adjustments, especially given the extreme 99.9% vs 0.1% imbalance.

Answer

Train the model using accuracy as the performance metric

Answer

Undersample the legitimate transactions to match the number of fraudulent ones

Answer

Increase the regularization strength in the model

Question 7

A company wants to deploy an AI model for real-time inference on edge devices with limited computational resources. Which model architecture would be MOST suitable?

Accepted Answer

MobileNet. MobileNet is specifically designed for mobile and edge devices using depthwise separable convolutions, which drastically reduce the number of parameters and computational cost while maintaining acceptable accuracy. This makes it the most suitable choice for real-time inference on resource-constrained edge hardware.

Answer

YOLOv4

Answer

ResNet-152

Answer

BERT

Question 8

A team is developing a natural language processing model to classify customer feedback. The dataset contains text in multiple languages. Which THREE preprocessing steps are essential to ensure the model performs well across all languages?

Accepted Answer

Lowercasing. Lowercasing (B) is essential because it normalizes case variations so that words like 'Good' and 'good' are treated as the same token, reducing vocabulary sparsity across all languages. Tokenization (C) is essential because it splits raw multilingual text into meaningful units (words, subwords, or characters), which is the foundational step required before any vectorization or model input. Removing stop words (E) is essential because high-frequency function words (e.g., 'the', 'and', 'de') add noise and carry little discriminative signal for sentiment/feedback classification, improving model focus across languages. One-hot encoding (A) is not a preprocessing step for multilingual text and produces sparse, high-dimensional vectors that do not generalize across languages. Stemming (D) is not essential and is language-dependent; it can even harm performance for morphologically rich or non-English languages, so it is not required to ensure cross-language performance.

Answer

One-hot encoding

Answer

Stemming

Question 9

A healthcare startup is developing a deep learning model to detect diabetic retinopathy from retinal images. The model is trained on a dataset of 10,000 labeled images. During initial testing, the model achieves 99% accuracy on the training set but only 85% on the test set. The startup wants to deploy the model in a clinical setting where false negatives (missing a disease) are critical. The team has access to additional unlabeled retinal images from multiple sources. Which strategy should the team use to improve the model's generalization and reduce false negatives?

Accepted Answer

Use semi-supervised learning with the unlabeled images to improve feature representations. Semi-supervised learning leverages the large pool of unlabeled retinal images to learn robust feature representations, which helps the model generalize better to unseen data. By reducing overfitting (the gap between 99% training and 85% test accuracy), this approach directly improves test-set performance. Additionally, semi-supervised methods can be tuned to emphasize recall, thereby reducing false negatives critical in clinical diabetic retinopathy screening.

Answer

Apply aggressive data augmentation to the training set

Answer

Increase the learning rate during training

Answer

Add more convolutional layers to the model

Question 10

A data scientist is preparing a dataset for training a classification model. The dataset contains 10,000 records with a binary target variable where 9,500 belong to class A and 500 belong to class B. Which technique should the scientist use to address the class imbalance?

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

Accepted Answer

Replace missing Age with the mean and missing Income with the median.. Using mean imputation for Age (10% missing) and median imputation for Income (30% missing) minimizes bias. Mean is suitable for roughly symmetric distributions with low missingness, while median is robust to outliers and skewness, which is common in income data. This combination reduces distortion of central tendency and preserves data integrity better than uniform methods.

Answer

Delete all rows with missing values.

Answer

Replace missing Age with the mode and missing Income with a constant value.

Answer

Replace missing values with zeros.

Question 11

A team is building a regression model to predict house prices. The dataset includes numerical features (square footage, number of bedrooms) and categorical features (neighborhood, roof type). The categorical features have high cardinality (neighborhood has 200+ unique values). Which encoding strategy should the team use to avoid overfitting and maintain model interpretability?

Accepted Answer

Target encoding with regularization.. Target encoding with regularization is the best choice because it replaces each categorical value with the mean of the target variable for that category, which captures the relationship between the category and house prices. Regularization (e.g., adding a prior or using cross-validation) shrinks the encoded values toward the global mean, preventing overfitting on rare categories (e.g., neighborhoods with only a few houses). This maintains interpretability because each encoded value directly reflects the average price impact of that category, unlike black-box embeddings.

Answer

Label encoding.

Answer

Binary encoding.

Answer

One-hot encoding with feature selection.

Question 12

A data engineer discovers that a dataset contains duplicate rows. Which data cleaning step is MOST appropriate?

Accepted Answer

Remove duplicate rows.. Removing duplicate rows is the most appropriate data cleaning step because duplicate rows can bias statistical analyses and machine learning models by overrepresenting certain observations. In data engineering, deduplication is a standard preprocessing step to ensure data integrity and avoid skewed results. Option C directly addresses this by eliminating redundant entries without introducing artificial values or altering the data distribution.

Answer

Keep only the first occurrence.

Answer

Fill duplicates with the mean.

Answer

Convert duplicates to categorical.

Question 13

A data scientist is working with a dataset that has 10,000 features but only 500 samples. The goal is to train a model for binary classification. Which feature selection technique is MOST appropriate to reduce overfitting?

Accepted Answer

Recursive Feature Elimination (RFE) with cross-validation.. Recursive Feature Elimination (RFE) with cross-validation is most appropriate because it recursively removes features and evaluates model performance, effectively handling high-dimensional data with few samples. Cross-validation helps mitigate overfitting by providing a more reliable estimate of feature importance. This method is robust for binary classification with many features.

Answer

Univariate selection using chi-squared test.

Answer

Using all features with L2 regularization.

Answer

Principal Component Analysis (PCA) without feature selection.

Question 14

A data scientist is cleaning a dataset. Which TWO actions are appropriate for handling missing data?

Accepted Answer

Use a predictive model to estimate missing values.. Option B is correct because using a predictive model to estimate missing values (model-based imputation, e.g., k-NN or regression imputation) leverages relationships between features to produce more plausible estimates than simple statistics, which is a legitimate and often superior technique for handling missing data. Option D is correct because when the missing rate is low, listwise deletion (dropping rows with missing values) removes only a small fraction of data and is unlikely to introduce substantial bias, making it a simple and acceptable approach. Option A is wrong because most training algorithms cannot process NaN values directly and ignoring missingness typically causes errors or biased results. Option C is wrong because imputing with the mean of the entire dataset ignores group structure and distorts variance and correlations, and it is inappropriate for categorical features. Option E is wrong because always replacing with the most frequent value (mode) is not universally valid—it is only suitable for categorical data and can distort distributions when applied indiscriminately.

Answer

Ignore missing values and train the model directly.

Answer

Impute missing values with the mean of the entire dataset.

Answer

Replace missing values with the most frequent value always.

Question 15

Which THREE data quality dimensions are critical for ensuring model reliability?

Accepted Answer

Timeliness.. Timeliness (A) is critical because stale or delayed data causes the model to learn and predict on outdated patterns, degrading reliability in production. Consistency (B) matters because conflicting values across sources or systems (e.g., the same customer having different attributes) introduce contradictory signals that destabilize training and inference. Accuracy (E) is essential since incorrect values directly corrupt the learned relationships, producing unreliable predictions regardless of other properties. Completeness (C) is not among the marked answers here, and Volume (D) is a data quantity/scalability concern rather than a core data quality dimension for model reliability.

Answer

Completeness.

Answer

Volume.