Geospatial modeling of landslide susceptibility in lateritic terrain: insights from machine learning techniques—random forest and DBSCAN

Abstract Landslides are a significant hazard in the Idukki district of Kerala, India, influenced by complex geological features, heavy rainfall, and steep terrain. This study aims to develop a predictive model for landslide susceptibility using machine learning techniques, integrating geological, hydrological, and topographical data. The conditioning factors analysed include geology, slope, rainfall patterns, soil texture, drainage density, lineament density, and soil moisture index. The Random Forest algorithm was employed to predict landslide-prone locations. The proposed framework, with an area under the curve (AUC) score of 0.95 in the receiver operating characteristic (ROC) curve, was used to map locations susceptible to rainfall-induced landslides in Idukki district. By integrating density clustering, the study classified the area into two categories: low and high landslide susceptibility. The majority of the study area falls under the high susceptibility category. The results were validated through cross-validation with historical landslide data, traditional overlay output, and density clustering classification, ensuring the accuracy and reliability of the predictions. The findings indicate that the proposed machine learning approach effectively identifies susceptible areas, offering essential insights for risk assessment and land-use planning. The study highlights the importance of avoiding rapid expansion of built-up areas and developmental activities in landslide-prone zones, underscoring the need for hazard-inclusive planning and land cover management for disaster risk reduction.