09. Study on the application of machine learning technique and geographical information systems to build landslide susceptibility maps

Abstract

Landslides have caused great damage to property, infrastructure and people in many mountainous areas of Vietnam. Studies on landslides have been addressed in the management and prevention of natural disasters. This study presents a machine learning method, which is a radial basis function classifier (RBFC) to create a landslide susceptibility map in Muong Cha district of Dien Bien province in the Northwest mountainous region, which is frequently affected by landslides. In the model, 12 influential factors were selected based on the topography and geographical conditions of the study area. To confirm the performance of the model, statistical indicators including ROC/AUC curves and various statistical indicators were used. The results showed that the RBFC models have high accuracy in building landslide spatial prediction maps, with AUC train = 0931, AUCtest = 0,857. This study is useful for building landslide susceptibility maps with the aim of identifying landslide prone areas for risk management.