Glossary
Updated
A combination of inherent (interpretable models) and post-hoc (SHAP, LIME, counterfactuals) methods that reveal feature importances, decision rules, or alternative outcome scenarios. Governance requires selecting explainability techniques suited to the model and audience, embedding explanations in user interfaces or compliance reports, and validating that explanations accurately reflect model behavior.
Example
A credit-card fraud model provides SHAP explanations with each alert: “Top factors: unusual location, atypical transaction size.” Fraud analysts use these explanations to triage alerts more effectively and regulators review the SHAP reports during compliance inspections.