Revision

Back to ML System Design

Feature Evaluation

1. Feature importance
2. Feature generalization

Feature importance

Model specific

Some models like tree models can output features importance. For example XGBoost:

Model agnostic

SHAP: SHapley Additive exPlanations

SHAP can be applied to measure a feature’s contribution to a single prediction:

Or to measure a feature’s contribution to the entire model:

Resources

See:

• Course on ML Interpretability,
• This Toward Data Science post.