First, we integrate ASD classification with SHAP-based connectivity interpretation. Second, we implement a leakage-controlled feature selection pipeline. Third, we map predictive features to candidate anatomical markers using the Harvard–Oxford atlas.
