Highway landslide susceptibility evaluation methods—a case study of Guangxi Zhuang autonomous region, China

Landslides pose a major threat to transportation infrastructure, especially in mountainous regions like the Guangxi Zhuang Autonomous Region. However, few studies have comprehensively compared susceptibility models for highway landslides in this region. To address this insufficiency, this study evaluates highway landslide susceptibility using 14 factors and seven models-three single models (frequency ratio, information value, certain factor) and four combined with frequency ratio (logistic regression, back propagation neural network, support vector machine, random forest). The findings indicate that all seven-evaluation models categorized the results into five susceptibility classes. The FR-BPNN yielded the highest percentage of very-low (28.79%) and low-susceptibility areas (49.84%), while the FR-SVM showed superior performance in identifying very-high-susceptibility zones (8.68%); Model analysis indicates elevated landslide susceptibility along highways in eastern Guangxi, particularly in Liuzhou and Guilin; ROC validation showed that all models achieved AUC > 0.80, with the FR-LR model attaining the highest accuracy (AUC = 0.91), outperforming traditional single-method models reported in previous studies. These results provide both a scientific tool for geological disaster prevention in the region and a reference framework for landslide risk assessment in areas with similar geographical conditions.