Undersampling Ensemble and Deep Learning - Based Landslide Susceptibility Mapping Method for Geological Hazard Warning

Landslide is the most frequent geological hazard. Landslide susceptibility mapping (LSM) can be used to predict the possibility of landslide occurring at a certain location. In this paper, an undersampling ensemble and deep learning - based landslide susceptibility mapping method for geological hazard warning is proposed. The Zigui to Badong section of the Three Gorges Reservoir is taken as the study area. Firstly, the correlation analysis of fourteen landslide influencing factors is carried out and two unimportant factors are eliminated. Then, an EasyEnsemble - one dimensional convolutional neural network (EE-1DCNN) model is constructed with the remaining twelve factors as inputs. Finally, the proposed EE-1DCNN model is compared with two well-known methods on test data, and a landslide susceptibility map of the study area is obtained based on the EE-1DCNN model. As the experimental result shows, the proposed EE-1DCNN model achieves superior AUC, accuracy and recall of 0.909, 91.7% and 85.1%. The applicability of the proposed method is proved.