Landslide Susceptibility Mapping in Complex Topo‐Climatic Himalayan Terrain, India Using Machine Learning Models: A Comparative Study of XGBoost, RF and ANN

ABSTRACT Landslides present a significant danger to both infrastructure and human lives in the challenging terrain of the Himalayas. Therefore, it is crucial to accurately map areas prone to landslides to facilitate informed decision‐making and proactive planning, allowing for effective management of this hazard. Since the landslide occurrences are accentuated by floods through toe erosion, and wildfires through this research aims to integrate machine learning techniques with the analysis of multiple hazards, such as floods and forest fires, as novel conditioning factors to create a comprehensive map of landslide susceptibility. Geospatial analysis was conducted to examine the relationship between 19 conditioning elements, including factors related to flood and forest fire susceptibility, which contribute to the occurrence of landslides. This study tested the efficacy of three machine learning models for mapping landslide‐prone areas: eXtreme Gradient Boost (XGBoost), Random Forest (RF) and Artificial Neural Network (ANN). These models can identify complex correlations and patterns among conditioning elements, resulting in more accurate mapping of regions prone to landslides. A regression analysis was performed to evaluate multicollinearity and confirm the association between the dependent and independent variables. The analysis revealed a variance inflation factor within acceptable bounds, providing validation for the correlation. The ROC–AUC curve approach was used to assess the models' accuracy. Among the models tested, XGB exhibited the highest accuracy at 94%, followed by RF at 92% and ANN at 77%. The results of this study offer insightful information about how to combine data from various hazard occurrences to forecast landslide susceptibility. This work can be instrumental for local authorities and disaster management organisations in prioritising resources, implementing mitigation plans and enhancing resilience against landslide threats.