Comparison of Different Machine Learning Models For Landslide Susceptibility Mapping

Abstract

The main objective of this study is to compare and evaluate three different machine learning models, namely logistic regression (LR), Naive Bayes (NB) and support vector machine (SVM), for landslide susceptibility mapping at the Jiuzhaigou area, Sichuan Province of China. A total number of 917 landslides visually interpreted from high resolution remotely sensed imaginaries and thirteen selected causing factors were used to model the landslide susceptibility for the study region. The success rates and prediction rates derived from Receiver Operating Characteristic curve analysis on the landslide susceptibility maps obtained from three models were investigated, the results suggested that the LR model outperformed the other two machine learning models, ranked in the top concerning general performance, followed by the NB model at the second, and the SVM model had the lowest accuracies, regarding to both success rates and prediction rates. This study provided a new perspective in selection of feasible model for landslide susceptibility mapping at the Jiuzhaigou area, Sichuan Province of China.