Community Detection on Model Explanation Graphs for Explainable AI

Ehsan Moradi

Feature-attribution methods (e.g., SHAP, LIME) explain individual predictions but often miss higher-order structure: sets of features that act in concert. We propose Modules of Inf…