This study focuses on landslide susceptibility prediction in Jiangxi Province by optimizing environmental factor discretization, using ChiMerge (CM), Natural Break (NB), and Equal Interval (EI) methods combined with statistical models (Frequency Ratio (FR), Certainty Factor (CF), and Information Value (IV)) and machine learning models (Random Forest (RF) and Support Vector Machine (SVM)). The results indicated that the CM method outperformed NB and EI, with higher FR, IV, CF peak values, enabling more effective discretization. The CM-FR and CM-RF models performed the best, achieving an AUC of 0.7944 and an overall accuracy of 74.93%, respectively. The high and very high susceptibility zones predicted by CM-RF model covered 15.24% of the total area, and included 88.12% of historical landslides, efficiently narrowing monitoring areas. The CM method provided the most rational discretization for environmental factors, significantly improving the accuracy and reliability of landslide susceptibility prediction compared with the NB and EI methods.
Li et al. (Fri,) studied this question.