AWS Certified Machine Learning – Specialty — Question 340

A data scientist needs to develop a model to detect fraud. The data scientist has less data for fraudulent transactions than for legitimate transactions.

The data scientist needs to check for bias in the model before finalizing the model. The data scientist needs to develop the model quickly.

Which solution will meet these requirements with the LEAST operational overhead?

Answer options

• A. Process and reduce bias by using the synthetic minority oversampling technique (SMOTE) in Amazon EMR. Use Amazon SageMaker Studio Classic to develop the model. Use Amazon Augmented Al (Amazon A2I) to check the model for bias before finalizing the model.
• B. Process and reduce bias by using the synthetic minority oversampling technique (SMOTE) in Amazon EMR. Use Amazon SageMaker Clarify to develop the model. Use Amazon Augmented AI (Amazon A2I) to check the model for bias before finalizing the model.
• C. Process and reduce bias by using the synthetic minority oversampling technique (SMOTE) in Amazon SageMaker Studio. Use Amazon SageMaker JumpStart to develop the model. Use Amazon SageMaker Clarify to check the model for bias before finalizing the model.
• D. Process and reduce bias by using an Amazon SageMaker Studio notebook. Use Amazon SageMaker JumpStart to develop the model. Use Amazon SageMaker Model Monitor to check the model for bias before finalizing the model.

Correct answer: C

Explanation

Option C is correct because running SMOTE directly in Amazon SageMaker Studio and using Amazon SageMaker JumpStart for model creation minimizes operational overhead compared to managing an Amazon EMR cluster. Furthermore, SageMaker Clarify is the appropriate tool to analyze model bias prior to finalization and deployment, whereas Amazon A2I is used for human-in-the-loop reviews and SageMaker Model Monitor is designed for detecting post-deployment drift in production.