Assessing landslide susceptibility based on hybrid Best-first decision tree with ensemble learning model

Landslide susceptibility mapping is a meaningful method to avoid and reduce the loss from landslide hazard. The main goal of current paper is to propose a hybrid model method to explore the effect of combining the Best-first decision tree (BFT) model with Bagging, Cascade generalization, Decorate, MultiboostAB, and Random SubSpace and measure the achievement of each combination model. Firstly, a landslide inventory map was produced using 364 landslides in the Yongxin County of China, then 364 non-landslide data were generated based on buffer method. Secondly, 255 landslides and 255 non-landslides were randomly chosen for the training data and the rest of 109 landslides and 109 non-landslides were chosen for validation data. Then, fifteen environment factors were chosen. Thirdly, the Support vector machines (SVM) method were applied to analysis the most useful factors for the modeling. The result demonstrated that all factors were useful for landslide modeling. Several statistical indexes were used to measure the performance, the results revealed that the five hybrid models performed better than the single BFT model. BFT-D and BFT-B were the best and effective models that can be adapted to model landslide susceptibility. The landslide susceptibility maps generated by the hybrid models will help land use arrangement and groundwork expansion in the Yongxin County.