GIS-Based Landslide Susceptibility Analysis Using Machine Learning Algorithms

GIS-based Landslide Susceptibility (LS) analysis of a selected Indian Himalayan region has been carried out using Gaussian Naïve Bayes, Logistic Regression, Radial Basis Function-Support Vector Machine, Decision Tree, and Random Forest algorithms. Hydro-geo-environmental landslide causative factors have been selected using statistical methods for training machine-learning models based on the selected algorithms. The performance of these algorithms is evaluated using statistical indices and the area under the ROC curve. Comparison and analysis of the trained models using the algorithms show that all the models perform satisfactorily for LS analysis (AUC ~ 0.82 – 0.97). The LS zonation maps generated from the trained models are validated using the landslide density method. The best algorithm for LS analysis and zonation mapping of the selected region is evaluated based on the algorithm performance evaluation, probability calibration analysis, and the quality of the generated zonation maps.