Novel Ensemble of M5P and Deep Learning Neural Network for Predicting Landslide Susceptibility: A Cross-Validation Approach

The landslides frequently affect Kurseong and the villages around it causing loss of life and property. The current study used the M5P technique, a deep learning neural network (DLNN), and an M5P-DLNN ensemble strategy to estimate the landslide susceptibility in the Kurseong area of West Bengal. The model’s results were cross-checked using the four folds of data. General field surveys and pertinent documents were used to find the locations of current landslides. Then, historical landslide locations were gathered, shown as an inventory map, and separated into four folds in order to calibrate and validate the models. A total of twelve LCFs (landslide conditioning factors) were employed to model the susceptibility to landslides. The developed landslide models were verified using two statistical techniques, namely the mean-absolute-error (MAE) and the root-mean-square-error (RMSE), as well as the receiver operating characteristic (ROC), accuracy, and precision. The accuracy measurements’ findings showed that all models had a good chance of identifying the Kurseong region’s landslide susceptibility. The ensemble model outscored the individual models in terms of precision among these models. The results of this study could be used to reduce the probability of landslides in the Kurseong region and other nearby locations with a similar topography and geology.