InSAR-Driven Negative Sample Purification Enhances Explainable Machine Learning for Landslide Susceptibility Map-ping: A Case Study of Ya’an, China

(1) landslide and non-landslide sample sets before (traditional buffer-based) and after InSAR-driven purification (GeoJSON/CSV format); (2) raw SHAP values and feature importance rankings of the optimized machine learning models (CSV format); (3) hyperparameters of the four models (Logistic Regression, Random Forest, Support Vector Machine, XGBoost) and raw calculation data of evaluation metrics (AUC, Accuracy, Precision, Recall, F1-Score) for six sampling strategy groups (Excel format); (4) nine static landslide causative factor layers selected by the Genetic Algorithm (GeoTIFF format).