Abstract: Benggang is one of the most severe types of soil erosion in the granite areas of southern China, due to the large erosion, strong explosiveness, and fast development speed. Accurate assessment of susceptibility is of great significance for the prevention and control of Benggang damages. In this study, different combinations of environmental factors and negative sample selection strategies were explored the impact on the assessment of Benggang susceptibility. A case study was taken from the granite area of Xingguo County, Ganzhou City, Jiangxi Province, China. A systematic detection was implemented to determine the explanatory power of 17 environmental factors on the development of Benggang using a GeoDetector (GD). According to the cumulative explanatory power percentage, 56.89%, 78.55%, 92.88%, and 100.00% were selected as the environmental factor combinations, corresponding to 4, 7, 10, and 17 environmental factors, respectively. Single random undersampling (SRU) was used to construct a negative sample dataset equal to positive samples using frequency ratio (FR). The susceptibility was calculated in the study area using automatic landslide susceptibility analysis (ALSA). Negative sample data was selected equal to positive samples in the low and extremely low susceptibility areas. The sample dataset was divided into the training and testing datasets in a 7:3 ratio. The training dataset was used to train the random forest (RF) model, and then the trained RF model was to calculate the testing dataset. The prediction accuracy of the model was evaluated to calculate the Benggang susceptibility using the receiver operating characteristic (ROC). The results show that: 1) The model accuracy under the three negative sample selection strategies decreased first and then increased with the increase of the number of factors. The area under curve (AUC) values of the model considering four environmental factors were 0.729, 0.909, and 0.909, respectively. The model accuracy was the lowest at 7 environmental factors, with the AUC values of 0.711, 0.869, and 0.893, respectively. The AUC values of the 10 environmental factors were 0.745, 0.942, and 0.919, respectively. The model accuracy was highest at 17 environmental factors, while the AUC values were 0.755, 0.947, and 0.929, respectively. There was the non-linear correlation between model accuracy and cumulative explanatory power percentage. The difference was only 0.020-0.038, although the accuracy of the model for 4 environmental factors was lower than that of 17 environmental factors. Therefore, the relatively ideal accuracy was achieved when considering the main controlling environmental factors; 2) The improved frequency ratio method was significantly improved the accuracy of the model. When the number of environmental factors was 4, the AUC values of FR and ALSA were both 0.909, and the negative samples selected by FR and ALSA were the most reasonable; When the number of environmental factors was 7, the AUC value of ALSA was 0.893, and the negative sample selected by ALSA was the most reasonable; When the environmental factors were 10 and 17, the AUC values of the FR were 0.942 and 0.947, respectively. In summary, the FR can be expected to select the most reasonable negative samples; 3) The acerage rainfall erosivity was closely related to the development of Benggang. Particularly, the acerage rainfall erosivity was ranged from 9 133.24 to 9 981.49 MJ∙mm/(hm²∙h∙a) within the scope of the study area. The majority of high and extremely high susceptibility areas were distributed in the southwest of the study area, whereas, a small number of extremely high susceptibility areas were distributed in the central and eastern parts of Xingguo County, and the majority of extremely low susceptibility areas were distributed in the northern and eastern. This study has explored the impact of different combinations of environmental factors and negative sample selection strategies on the susceptibility assessment of Benggang. The finding can provide the scientific basis for the disaster prevention and reduction in granite areas.