ABSTRACT
The main urban area of Kangding City and its surrounding watershed were taken as the study area. Sixteen environmental factors that can potentially cause landslides were identified and selected. The Apriori algorithm was used to screen out the combination of environmental factors. Put this combination into a random forest (RF) model for prediction. The results of the factor-screened model were compared with that of the original random forest model. The comparison of the results evaluated by frequency ratio accuracy and receiver operating characteristic curve (ROC) shows that the prediction results of the random forest model after factor screening are highly consistent with the actual landslide hazard distribution, and can make better predictions in the spatial distribution of landslide susceptibility of the study area. This study provides a more appropriate method for eliminating redundancy and noise factors in the environmental factors of landslide disasters.
