Comparative Analysis of Machine Learning, Statistical, and MCDA Methods for Rainfall-Induced Landslide Susceptibility Mapping in the Eco-Sensitive Koyna River Basin of India

This study aimed to identify the most effective method for landslide susceptibility mapping (LSM) in the Koyna River Basin, a region in the Western Ghats that is prone to devastating landslides. The study used a comparative analysis of three methods: support vector machine (SVM), frequency ratio (FR), and analytical hierarchy process (AHP) to determine the most effective method for LSM. A total of 1823 landslide events were identified and divided into training and test datasets. The study mapped various layers of landslide influencing factors and selected the optimal subset of factors using the information gain ratio. The susceptibility of landslides was mapped using SVM, FR, and AHP models, and the accuracies of all models were evaluated using the test dataset. The study showed that the SVM model was the most effective method with an area under the curve (AUC) value of 0.87, indicating acceptable prediction effectiveness. The LSM result of SVM was reliable and effective for implementation. The result was showing high landslide susceptibility in the forest area, particularly in the region of the Koyna Wildlife Sanctuary. The study provides valuable insights into effective methods for LSM in landslide-prone regions, which can help minimize losses and improve disaster management strategies.