An Advanced Hybrid Machine Learning Technique for Assessing the Susceptibility to Landslides in the Upper Meenachil River Basin of Kerala, India

The ambition of the current study was to generate landslide susceptibility maps (LSMs) for the Meenachil river basin’s upper catchment using the ensemble NBT-RTF, Naive Bayes tree (NBT), and rotation forest (RTF). For landslide susceptibility modelling, 189 landslide sites and 12 landslide conditioning factors (LCFs) were gathered. Multi-collinearity analysis was done among the LCFs to determine the best LCFs to use. The metrics utilized to assess the predictive power of the employed models are ROC-AUC, mean-absolute-error (MAE), root-mean-square-error (RMSE), and kappa coefficient. Almost 14% of the studied region has very high landslide susceptibility, according to the results of the best-performed model. The NBT-RTF model got the lowest RMSE and the greatest ROC-AUC (0.867) and kappa index (0.884) during the validation phase (0.234). The anticipated model is reliable for minimizing the impact of landslides in the research region and planning land development.