Spatial prediction of landslide susceptibility based on the neighborhood rough set

Abstract

This paper discusses the feasibility of reducing the landslide inducing factors by the neighborhood rough set algorithm in data processing section, which could improve the accuracy and timeliness of landslides susceptibility prediction models effectively. 15 predisposing factors for a continuous value that has not been graded were reduced by nighborhood rough set, a granularity calculation method, based on the importance degree of each factor. Then the combination of factors before and after optimization was put into random forest (RF) and support vector machine (SVM) for modelling. ROC curve and statistical indicators show that: the average performance of the reduced factors combination is superior to that before optimization. In addition, we used the RF which has a better performs in evaluation to map the landslides susceptibility in Jiuzhaigou area, discuss the timeliness of the assessment of landslides prediction and the weight of the predisposing factors.