Question

A data scientist has trained an XGBoost model in SageMaker to approve or deny small-business loan applications. Compliance officers require an explanation of which input features most influenced each individual prediction, so that adverse-decision notices can be sent to applicants. The team wants a solution that produces per-instance feature attributions using a widely accepted, model-agnostic method. Which approach best meets this requirement?

Accepted Answer

Run a SageMaker Clarify processing job configured for SHAP explainability to generate local (per-instance) feature attributions.

Answer

Enable SageMaker Model Monitor data-quality monitoring to capture feature statistics for each prediction.

Answer

Retrieve the global feature_importances_ attribute from the trained XGBoost booster and report it for every applicant.

Answer

Use SageMaker Autopilot to regenerate the model and read the leaderboard candidate metrics.