Explainable AI Integrated Feature Selection for Landslide Susceptibility Mapping Using TreeSHAP

Landslides have been a regular occurrence and an alarming threat to human life property in the era of anthropogenic global warming. An early prediction landslide susceptibility using data-driven approach is demand time. In this study, we explored eloquent features that best describe with state-of-the-art machine learning methods. our employed algorithms including XgBoost, LR, KNN, SVM, Adaboost for prediction. To find hyperparameters each individual classifier optimized performance, incorporated Grid Search method, 10 Fold Cross-Validation. context, version XgBoost outperformed all other classifiers Cross-validation Weighted F1 score 94.62 %. Followed by empirical evidence, incorporating TreeSHAP, game-theory-based statistical algorithm used explain Machine Learning models, identify such as SLOPE, ELEVATION, TWI complement performance XGBoost mostly LANDUSE, NDVI, SPI which has less effect on models performance. According TreeSHAP explanation features, selected 9 most significant causal factors out 15. Evidently, along feature reduction 40 % terms popular evaluation metrics Cross-Validation 95.01 training AUC 97