This study addresses the susceptibility assessment of landslide disasters on transmission line tower slopes in the Longnan region, employing four machine learning models—KNN, LR, decision tree, and XGBoost—based on a systematic analysis of 10 influencing factors. Initially, the influencing factor system was optimized through correlation analysis, revealing significant correlations among topographic factors. The exclusion of the roughness factor notably enhanced the statistical performance of the models. Regional evaluation results indicate a spatial heterogeneity in landslide susceptibility, characterized by low susceptibility in the northwest and high susceptibility in the southeast. Extremely high-susceptibility zones are predominantly distributed in areas with intense human-land interactions, such as active fault zones, deeply incised valleys, and along major transportation corridors. Comparative analysis of the models demonstrates that the KNN model exhibits systematic misclassification, while the LR and decision tree models show insufficient recognition rates for high-slope hazard zones. The XGBoost model outperforms the others overall but still presents limitations under complex geological conditions. The findings provide a scientific basis for landslide prevention and mitigation on transmission line slopes and highlight directions for model optimization.