GIS and remote sensing-based landslide susceptibility mapping in Phuket using machine learning with feature selections

Phuket’s high susceptibility to landslides is driven by its steep terrain, heavy rainfall, and rapid, tourism-fueled urbanization. This study assessed landslide risk using GIS, remote sensing together with logistic regression (LR), support vector machine (SVM), and Neural Networks (NN). We initially analyzed twelve factors, reducing them with two methods: an optimization feature selection method and a statistical correlation test. Both methods delivered strong, comparable predictive performance, with a precision of 88-90% and an AUC of 0.93-0.95. The Neural Network model proved most effective, with a prediction error under 3%, demonstrating its ability to capture complex landslide patterns. The final susceptibility map highlights high-risk zones in Kathu, northern Mueang Phuket, and eastern Thalang. While this research showcases the power of machine learning, the map’s ultimate effectiveness relies on its integration into a governance framework that enforces land-use regulations and supports proactive disaster management.