Abstract: Landslide susceptibility assessment is of great significance for landslide risk management and land-use planning. This study focuses on the main urban area of Badong County, collecting data from 50 historical landslides and constructing a susceptibility index system comprising 14 influencing factors. To address the limitations of traditional single machine learning models, we proposed a hybrid model based on Bayesian optimization for a comprehensive evaluation of landslide susceptibility. The model was further interpreted using the SHapley Additive exPlanations(SHAP)method. The results indicate that the BO-XGBoost model achieved an Area Under the Curve(AUC) of 0.982, significantly higher than the scores of the BO-LR, BO-DT, BO-SVM, BO-RF, and BO-LGBM models, demonstrating strong predictive performance of the BO-XGBoost model. The landslide susceptibility map revealed that high-susceptibility areas in the main urban area of Badong County were primarily concentrated along riverbanks and in regions with dense road networks, while low-risk areas are predominantly found in high-elevation, densely vegetated zones. Frequency ratio analysis confirmed that the susceptibility map effectively distinguished between high and low susceptibility areas. An in-depth explanation based on SHAP identified key factors influencing landslide occurrences in the study area, including distance to streams, annual rainfall, land use type, and distance to faults, while providing insights into the relative contributions of each factor to susceptibility during individual predictions. This research offers new perspectives on landslide susceptibility modeling and its interpretability.