Abstract: Landslides are among the most frequent and destructive geological hazards in mountainous areas of southwestern China, and their susceptibility is jointly controlled by multiple factors including topography, geology, meteorology, and human activities. To improve the accuracy of landslide risk identification in complex mountainous areas, this study takes Dechang County, Sichuan Province as the research area, and establishes three landslide susceptibility evaluation models: Logistic Regression (LR), Random Forest (RF), and Gradient Boosting Decision Tree (GBDT) based on typical landslide influencing factors. A systematic comparative analysis is carried out from the perspectives of prediction performance, spatial distribution characteristics, and identification of dominant controlling factors. Nine influencing factors are selected for modelling, and the Area Under the ROC Curve (AUC), Kappa coefficient, and Overall Accuracy (ACC) are adopted to assess model performance. Results show that all three models can effectively reflect the spatial distribution pattern of landslide susceptibility in Dechang County. The GBDT model achieves the highest prediction accuracy and strong capability in identifying extremely low susceptibility zones; the RF model performs stably with high generalization ability; the LR model has good interpretability but relatively low prediction accuracy due to multicollinearity and linear hypothesis constraints. In terms of feature importance, the Normalized Difference Vegetation Index (NDVI), slope, and distance to roads are the dominant factors identified by RF and GBDT models, which reflects the disaster-causing mechanism of landslides in Dechang County under the dual effects of vegetation coverage and engineering disturbance. Under the sample size and factor configuration of this study, ensemble learning models are overall superior to traditional statistical models in terms of landslide susceptibility prediction accuracy and stability, verifying the effectiveness and applicability of ensemble learning–based susceptibility assessment in complex mountainous environments. The model evaluation framework and technical process established in this study provide a reference for model selection and optimization of landslide susceptibility assessment, and support geological hazard risk identification, territorial spatial planning, and engineering site selection in Dechang County and similar mountainous areas.