Abstract: Landslides are among the most frequent and destructive geological hazards, characterized by significant spatial heterogeneity and complexity. Aiming to enhance the accuracy and interpretability of Landslide Susceptibility Mapping, this study, based on landslide data from Xianhe Town, Ankang City, selected ten evaluation factors, including slope structure, slope type, geomorphology, plan curvature, profile curvature, aspect, slope, elevation, road buffer zone, and river buffer zone. Four Machine Learning models, including Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), and Light Gradient Boosting Machine (LightGBM), were employed to evaluate landslide susceptibility. Model performance was systematically assessed using metrics such as AUC, Accuracy, Precision, Recall, F1-Score, and MCC. Furthermore, the rationality of the models was validated by analyzing the distribution of historical landslides within different susceptibility zones. The SHAP (SHapley Additive exPlanations) approach were integrated to unveil the underlying influence mechanisms of the conditioning factors. The results indicate that the RF model yielded the optimal performance (AUC=0.8413). The very high susceptibility zone delineated by this model accounts for only 6.79% of the total area but captures 47.17% of historical landslides, demonstrating exceptional zoning efficiency. The SHAP analysis identifies the road buffer zone is a critical factor controlling landslide susceptibility. Additionally, a negative correlation was observed between plan curvature and profile curvature. This study provides a scientific basis for regional landslide prevention and spatial planning.