GIS-based landslide hazard zonation and risk mapping along NH-9 in the Himalaya using ensemble machine learning and ANN models

Abstract This study evaluates and compares the performance of machine learning (ML) algorithms, namely Random Forest (RF), Logistic Regression (LR), Decision Tree (DT) and Multilayer Perceptron (MLP) for delineating high-resolution landslide hazard and risk maps along a section of National Highway-9 (NH-9) in Uttarakhand, India. A total of fifteen landslide conditioning factors (LCFs) were selected with special emphasis on field based geotechnical and bedding orientation relative to slope were incorporated. Multicollinearity analysis verified variable independence and reliability and the models were trained on a balanced dataset of 95 landslide and non-landslide locations. The limited spatial extent of high and very high hazard zones indicates localized instability, implying that risk mitigation efforts can be strategically planned and focused on discrete vulnerable sections. Model performance was assessed through Receiver Operating Characteristic (ROC) curve, Precision-Recall, accuracy, F1 score and Kappa coefficient. The ROC analysis indicated that all four ML models achieved excellent discrimination (AUC > 0.90) along with validation from other metrices. Among them, the RF model exhibited the best overall performance, with an AUC of 96.8% for the training dataset (Success Rate Curve) and 96.0% for the testing dataset (Prediction Rate Curve), indicating excellent predictive capability and strong generalization. Although the DT model achieved the second highest testing AUC (95.1%), its lower training AUC (94.1%) suggests slightly less consistent performance across datasets. The MLP and LR models also demonstrated good predictive performance but were marginally inferior to the RF and DT models. This study further extends the applicability of landslide hazard outputs by generating landslide risk maps through the integration of elements at risk and their vulnerability. Population and LULC were utilized as exposure elements for risk zonation mapping. The outputs highlight the effectiveness of ensemble ML techniques in landslide studies providing insights for hazard mitigation, infrastructure planning and sustainable development along this Himalayan transportation corridor.