Landslide Susceptibility Mapping in Guangdong Province, China, Using Random Forest Model and Considering Sample Type and Balance

Landslides pose a serious threat to human lives and property. Accurate landslide susceptibility mapping (LSM) is crucial for sustainable development. Machine learning has recently become an important means of LSM. However, the accuracy machine models limited by heterogeneity environmental factors imbalance samples, especially large-scale To address these problems, we created improved random forest (RF)-based LSM model applied it Guangdong Province, China. First, RF-based was constructed using rainfall-induced samples 13 exploring optimal positive-to-negative training-to-test sample ratios. Second, performance evaluated compared with three other models. The results indicate that: (1) proposed best highest area under curve (AUC) 0.9145, based on ratios 1:1 8:2, respectively; (2) introduction rainfall global modification (GHM) can increase AUC from 0.8808 0.9145; (3) topography are two dominant in landslides. These findings facilitate risk prevention serve as technical reference accurate