Performance of logistic regression and support vector machine conjunction with the GIS and RS in the landslide susceptibility assessment: Case study in Nakhon Si Thammarat, southern Thailand

Abstract

The occurrence of landslides has risen in the past few decades, particularly in mountainous regions worldwide, including Nakhon Si Thammarat, southern Thailand. Despite various methods being employed for the initial management of landslide disasters, none have proven universally effective. The goal of this research is to create and assess landslide susceptibility maps (LSMs) within this area by employing support vector machine (SVM) and logistic regression, together with Geographic Information System (GIS) and Remote Sensing (RS) techniques. Eleven factors contributing to landslides were identified as topographic, environmental, and geological influences. The 365 landslides in the past were aimlessly selected into training (70%) and testing (30%) datasets. The four LSMs indicated that approximately 13%–20% of this study area exhibit a high susceptibility to landslides corresponding to the regions of high elevation with relatively steep slope angles. To evaluate and compare LSM models, the AUC value for training dataset were 0.977, 0.975, 0.958, and 0.967 and testing dataset were 0.973, 0.969, 0.956, and 0.964 for SVM with the radial basis function (rbf) kernel, SVM with polynomial deg 2, SVM with linear kernel and logistic regression models, respectively. Among these models, SVMs with rbf demonstrated the highest prediction rate. However, it requires a significant amount of time to choose the best parameters for achieving the highest accuracy prediction. In summary, these maps are applicable at the regional level to enhance the management of landslide hazards.

Keywords

• Landslide susceptibility maps (LSMs)
• Logistic regression
• Support vector machine (SVM)
• Geographic Information System (GIS)
• AUC