explainability_notes.md

🔍 Explainability Layer Documentation

Dual-Approach Interpretability System

SHAP (SHapley Additive exPlanations)

Purpose: Global model interpretability across entire dataset Implementation: TreeExplainer for XGBoost models Key Insights:

• Top features: \ca\ (number of major vessels), \ hal\ (thalassemia), \cp\ (chest pain type)
• Global feature importance rankings
• Expected value: -0.243 (base prediction)

Visualizations:

• Summary plots for feature importance
• Force plots for individual predictions
• Dependence plots for feature relationships

LIME (Local Interpretable Model-agnostic Explanations)

Purpose: Local interpretability for individual predictions Implementation: TabularExplainer for per-instance explanations Key Benefits:

• Case-by-case medical reasoning
• Trust building with clinicians
• Regulatory compliance support

Clinical Validation

• Coverage: 100% of predictions explained
• Latency: <200ms per explanation
• Accuracy: Local surrogate model fidelity >92%