Artificial Intelligence and Machine Learning Driven Landslide Analysis: Scar Segmentation, Susceptibility and Sensitivity Analysis using Naive Bayes and K-nearest neighbor in Nainital District, Uttarakhand
Landslides represent a recurrent hazard in mountainous regions, notably in the geologically fragile Nainital district of Uttarakhand, India. This study develops a data-driven landslide susceptibility model using k-Nearest Neighbors (KNN) and Naive Bayes (NB) classifiers, integrating 273 inventoried landslide scars and eight geo-environmental conditioning factors. The dataset was partitioned into 70% training and 30% testing subsets. Model performance, evaluated using area under the ROC curve (AUC), yielded predictive accuracies of 0.79 (KNN) and 0.81 (NB). The analysis revealed that 33.92% of the study area falls within high to very high susceptibility zones, concentrated along south- and southwest-facing slopes (25°-60°), elevations of 900-1800 m, and areas proximal to roads and drainage. Stream Power Index and Topographic Wetness Index emerged as dominant contributors. The findings affirm the applicability of AI-based models for reliable landslide prediction and offer strategic insights for risk-informed planning and disaster mitigation in Himalayan terrain.
