Explainable machine learning for landslide susceptibility assessment incorporating InSAR deformation rates

To enhance landslide susceptibility assessment in southeastern Chongqing, this study develops an interpretable machine learning framework integrating SBAS-InSAR dynamic deformation rates. Positive samples integrate historical records with newly detected landslides, balanced by Information Value-selected negative samples. Seven models are optimized via the Sparrow Search Algorithm (SSA) and interpreted using SHAP. Experimental results demonstrate that the SSA-XGBoost model performs best (AUC = 0.9307), delineating high-susceptibility zones that cover 12.878% of the area yet capture 61.690% of landslides. Elevation, NDVI, deformation rate, and proximity to drainage and transportation networks constitute the primary drivers.