A finance team has stock return data for 5,000 companies with 2,000 features each. They want to identify the top 15 features most useful for predicting future returns with minimal operational overhead. Which approach meets this need?

SageMaker Autopilot automates model training and hyperparameter tuning, significantly reducing operational overhead. It automatically generates multiple models and provides a leader board. After a model is trained, SageMaker Clarify can be used to generate feature importance reports (e.g., SHAP values), which are robust and model-agnostic, fulfilling the requirement to identify the most useful features. Training a linear regression model and ranking by absolute coefficient values is a valid feature selection method but might not capture non-linear relationships and requires manual model building. Random Forest Gini importance is also a valid method but still requires manual model building and might be biased towards high-cardinality features. SageMaker Data Wrangler's quick model visualization is for exploratory analysis and might not provide the robust, production-ready feature importance needed for critical predictions.