This research aims to identify key factors for landslide susceptibility mapping using interpretable machine learning methods. It focuses on efficient factor selection in the Three Gorges Reservoir Area.
Selected 25 evaluation factors based on topography, geology, and hydrology.
Employed machine learning models RF, DT, and XGBoost for mapping susceptibility.
Utilized SHAP and LIME for model interpretation and compared results with AUC-RFE.
Identified a core factor set with AUC value 0.931—0.003 lower than 13 filtered factors.
Core factor sets from interpretable methods outperformed those from AUC-RFE.
Achieved a significant improvement in the efficiency of factor selection for landslide susceptibility.