Question 1
A data scientist is building a classification model and notices that the training accuracy is very high, but the test accuracy is very low. What is the most likely issue?
A. Underfitting
B. Overfitting
C. Proper generalization
D. Data leakage

Correct Answer: B
Explanation: A high training accuracy and low test accuracy usually indicate overfitting, where the model memorizes the training data instead of learning generalizable patterns.

Question 2
You are training a deep learning model on SageMaker. The training job is taking too long due to a large dataset. What can you do to improve performance?
A. Use a smaller instance
B. Use SageMaker Debugger
C. Enable pipe input mode
D. Split the dataset into multiple files

Correct Answer: C
Explanation: Pipe input mode allows SageMaker to stream data directly from Amazon S3 to the training instances, reducing startup time and speeding up training with large datasets.

Question 3
Which AWS service can be used to prepare, transform, and clean data for ML workflows in a serverless environment?
A. Amazon EMR
B. Amazon Athena
C. AWS Glue
D. AWS Lambda

Correct Answer: C
Explanation: AWS Glue is a serverless ETL service designed to clean, prepare, and transform data for analytics and ML workflows.

Question 4
What metric is best suited to evaluate a binary classifier when dealing with imbalanced classes?
A. Accuracy
B. Precision
C. Recall
D. F1 Score

Correct Answer: D
Explanation: The F1 score balances both precision and recall, making it suitable for imbalanced datasets where accuracy alone can be misleading.

Question 5
Which Amazon SageMaker feature lets you automatically search for the best combination of hyperparameters?
A. SageMaker Ground Truth
B. SageMaker Autopilot
C. SageMaker Tuning
D. SageMaker Clarify

Correct Answer: C
Explanation: SageMaker Automatic Model Tuning (also known as Hyperparameter Tuning) automatically searches for the best model parameters.

Question 6
Which AWS service can be used to label training datasets efficiently using human workers?
A. Amazon SageMaker Ground Truth
B. Amazon Mechanical Turk
C. AWS DataBrew
D. AWS Comprehend

Correct Answer: A
Explanation: SageMaker Ground Truth provides a scalable way to label datasets using humans and machine learning models.

Question 7
Which method can prevent overfitting in a deep learning model?
A. Adding more layers
B. Increasing batch size
C. Dropout regularization
D. Using a lower learning rate

Correct Answer: C
Explanation: Dropout randomly turns off neurons during training, helping the model generalize better and reducing overfitting.

Question 8
Which AWS service is purpose-built to run and deploy trained ML models in production at scale?
A. Amazon SageMaker Studio
B. Amazon SageMaker Hosting Services
C. Amazon EC2
D. Amazon Athena

Correct Answer: B
Explanation: SageMaker Hosting Services enables the deployment of trained models for real-time or batch inference.

Question 9
A machine learning team wants to ensure their model behaves fairly across different groups. Which SageMaker tool can help?
A. SageMaker Clarify
B. SageMaker Autopilot
C. SageMaker Debugger
D. SageMaker Model Monitor

Correct Answer: A
Explanation: SageMaker Clarify detects bias in data and models to support fairness and transparency in ML.

Question 10
What is the primary benefit of using Amazon SageMaker Autopilot?
A. Manual control over every modeling step
B. Automatic model deployment with no user input
C. Automated machine learning (AutoML) that still provides transparency into model details
D. Integration with AWS Snowball

Correct Answer: C
Explanation: SageMaker Autopilot automates the entire machine learning workflow while allowing visibility into how models are built and selected.

Question 11
A data science team wants to deploy a trained model for real-time inference. Which SageMaker option is best suited for this?
A. SageMaker Batch Transform
B. SageMaker Ground Truth
C. SageMaker Hosting Endpoint
D. SageMaker Pipelines

Correct Answer: C
Explanation: SageMaker Hosting Endpoint provides a persistent HTTP endpoint for real-time inference.

Question 12
Which evaluation metric is most appropriate for a multi-class classification problem?
A. ROC AUC
B. Mean Absolute Error
C. Confusion Matrix
D. R² Score

Correct Answer: C
Explanation: A confusion matrix is widely used for evaluating multi-class classification models.

Question 13
Which method is commonly used to standardize features in a dataset for ML modeling?
A. One-hot encoding
B. Min-max scaling
C. Label encoding
D. Log transformation

Correct Answer: B
Explanation: Min-max scaling brings all feature values into a common scale (e.g., 0 to 1), which is important for many ML algorithms.

Question 14
A developer wants to automatically retrain a model whenever new training data is added to S3. Which AWS service combination is best for this task?
A. AWS Lambda + SageMaker Training + Amazon S3
B. Amazon SNS + Amazon EC2 + SageMaker
C. Amazon SQS + AWS Glue + SageMaker
D. Amazon CloudTrail + SageMaker + Amazon RDS

Correct Answer: A
Explanation: AWS Lambda can trigger retraining when new data is added to Amazon S3, and SageMaker Training Jobs can be used to retrain the model.

Question 15
What is data leakage in machine learning?
A. When training data is lost during preprocessing
B. When validation data is included in training data
C. When irrelevant features are used for training
D. When output labels are not encoded properly

Correct Answer: B
Explanation: Data leakage occurs when information from the test or validation set unintentionally influences the training process, resulting in overly optimistic performance.

Question 16
Which AWS service allows you to orchestrate end-to-end ML workflows using steps like processing, training, and deployment?
A. Amazon Step Functions
B. AWS Glue
C. SageMaker Pipelines
D. AWS CodePipeline

Correct Answer: C
Explanation: SageMaker Pipelines provides a native way to define, visualize, and manage end-to-end ML workflows.

Question 17
Which of the following would improve model generalization?
A. Reduce dataset size
B. Increase model complexity
C. Use cross-validation
D. Over-sample the dominant class

Correct Answer: C
Explanation: Cross-validation helps evaluate how well a model generalizes to unseen data and is a good technique to avoid overfitting.

Question 18
A data scientist needs to reduce the number of features in a dataset while keeping most of the variance. Which technique should be used?
A. Label encoding
B. L2 regularization
C. PCA (Principal Component Analysis)
D. Random forest

Correct Answer: C
Explanation: PCA is a dimensionality reduction technique that preserves as much data variance as possible.

Question 19
Which SageMaker feature allows you to monitor drift in model predictions after deployment?
A. SageMaker Clarify
B. SageMaker Monitor
C. SageMaker Debugger
D. SageMaker JumpStart

Correct Answer: B
Explanation: SageMaker Model Monitor detects data and model drift to help ensure prediction quality over time.

Question 20
Which AWS service allows you to run interactive data analysis and build Jupyter notebooks?
A. AWS Lambda
B. Amazon Athena
C. Amazon SageMaker Studio
D. AWS Glue

Correct Answer: C
Explanation: SageMaker Studio is a web-based IDE that provides Jupyter notebooks for building, training, and deploying ML models interactively.