Artificial Neural Network Ensemble with General Linear Model for Modeling the Landslide Susceptibility in Mirik Region of West Bengal, India

Landslide is one of the important problems in the Mirik region of West Bengal. For managing this problem it is important to delineate the areas which are highly susceptible to landslide. In the present study ensemble of ANN, general linear model (GLM), and ensemble ANN-GLM machine learning methods were applied for producing the landslide susceptibility maps (LSMs) of the Mirik region. A total of 373 landslide locations and twelve landslide conditioning factors (LCFs) are retrieved from the spatial database and used for modeling the landslide susceptibility. Multicollinearity between the LCFs was carried out in order to select suitable LCFs. The built-in models were validated using ROC-AUC, mean absolute error (MAE), root mean square error (RMSE), and kappa coefficient. Using the 70:30 ratio landslide locations were classified into training and testing datasets. The ANN-GLM model got the lowest RMSE and the highest ROC-AUC (0.864) and kappa index (0.889) during the validation phase (0.086). As per the result of ensemble model 20.99% area of the Mirik region is very highly susceptible for landslide. The anticipated model is reliable in lowering the danger of landslide risks for prospective land use planning in the Mirik region of West Bengal 112.