Improving Landslide Susceptibility Prediction in Uttarakhand through Hyper-Tuned Artificial Intelligence and Global Sensitivity Analysis

Landslides are constantly increasing in the Himalayan region due to strong tectonic activities, soil erosion, heavy rainfall, and anthropogenic activities. Despite the severe impact of landslides in the Himalayan regions on human lives and property, there is an urgent need for research to assess landslide vulnerability map and make better decisions. Therefore, this study aims to predict an accurate landslide susceptibility using hyper-tuned random forest (RF) and deep neural network (DNN) in the Himalayan State of Uttarakhand. Sobol’s global sensitivity and Explainable Artificial Intelligence (XAI) were also used to evaluate the impact of sixteen priority landslide conditioning parameters (LCPs). The study shows that, in both models, the areas with “very high” and “high” susceptibility to landslides are 5.72–6.29% and 17.33–17.37% of the total area, respectively. Furthermore, the Receiver Operating Characteristic (ROC) curve and Precision-Recall (PR) curve show that the Random Forest (RF) model outperforms the Deep Neural Network (DNN) model with an AUC value of 0.95 and 0.95, respectively. Sobol’s sensitivity analysis for RF shows that elevation, annual rainfall and slope are the most influential parameters in LSM. The study presents an innovative approach that combines hyper-tuned ML and DL with XAI techniques to create a robust model for LSM that can be used in other regions of the world to assess susceptibility landslide.