Landslides cause significant economic losses and human casualties. Landslide susceptibility assessment (LSA) is essential for land-use planning and risk management. This study aims to investigates hybrid models that fully leverages the advantages of individual supervised learning (ISL) models and ensemble learning (EL) models. We compared the performance of six supervised learning algorithms—logistic regression (LR), support vector machine (SVM), random forest (RF), adaptive boosting (AdaBoost), gradient boosting decision tree (GBDT), and extreme gradient boosting (XGBoost)—for LSA in Qiongzhong County, Hainan Island. Initially, twelve conditioning factors were quantified using certainty factors (CF) and filtered via independence tests. Subsequently, the landslide susceptibility maps (LSMs) were generated using the six models. Lastly, the model performance and LSMs were evaluated using statistical metrics. Results indicate that EL models, particularly XGBoost (AUC = 0.847) and RF (AUC = 0.843), outperformed ISL models. The normalized difference vegetation index (NDVI) was the most influential factor, followed by distance to roads. This study demonstrates the superiority of EL in enhancing prediction accuracy, addressing a research gap in LSA model comparison for tropical granite regions, and providing a reliable approach for disaster management.
Li et al. (2026) studied this question.