Analysis of Landslide Susceptibility in Qingyuan City Based on Machine Learning and SINMAP Coupling Model

A variety of complex geological disasters induced by extreme rainfall seriously threaten people's production and life safety, among which landslide disasters are the most widely distributed and the most destructive. In this paper, the typical rainfall concentration area of Guangdong Province (Qingxin-Qingcheng-Fogang-Yingde) is taken as the research area. Based on multi-source data, four machine learning methods are used to calculate the vulnerability of geological disasters such as landslides induced by heavy rain and flood, and the importance of the characteristic factors is analyzed to complete the landslide hazard susceptibility map. On this basis, the SINMAP model is coupled to enrich the data samples, improve the generalization ability of the machine learning model, and provide theoretical support for the risk prevention and resistance in the disaster situation of Guangdong Province. The results show that XGBoost model has the best performance, and its AUC value is 0.90, which is the largest among the four machine learning models. Among the 16 features, elevation contributes the most to the occurrence of landslide disasters, accounting for 15%; In XGBoost-SINMAP coupling model, the accuracy rate is 89%, the precision rate is 87%, and the recall rate is 92%, which further improves the prediction performance of XGBoost model.