Evaluation of Machine Learning Algorithm for Landslide Sensitive Spatial Model Based on GIS Taking the Area along Sichuan-Tibet Railway as an Example

Abstract

With the development of GIS-based geospatial information technology, new methods have been provided for landslide disaster research. However, there is currently no suitable way to combine GIS technology with machine algorithms. In order to construct a highly accurate landslide sensitivity spatial model, this paper statistically analyzes the relationship between landslide disasters and various influencing factors in the study area through the GIS spatial analysis function, especially for the actual situation along the Sichuan-Tibet Railway. In this paper, a cross-check method is used to construct a landslide sensitivity evaluation model, and the accuracy of different models is quantitatively evaluated. The results of the fitting accuracy of the logistic regression model and the support vector machine model are: the average accuracy in the modeling stage is 75.722 and 75.65, and the average accuracy in the verification stage is 71.34 and 71.21. At the modeling stage, the SVM model has a fitting accuracy of about 3% higher than that of the logistic regression model; at the verification stage, the fitting accuracy is 0.13% higher than that of the logistic regression model; the AUC results show that the SVM model performs optimally, its AUC value is above 0.9, which achieves a higher accuracy. Compared with the logistic regression model, this value is 0.111 higher in the modeling stage and 0.111 higher in the verification stage.