Landslide susceptibility modeling using Reduced Error Pruning Trees and different ensemble techniques: Hybrid machine learning approaches

Abstract

Nowadays, a number of machine learning prediction methods are being applied in the field of landslide susceptibility modeling of the large area especially in the difficult hilly terrain. In the present study, hybrid machine learning approaches of Reduced Error Pruning Trees (REPT) and different ensemble techniques were used for the construction of four novel hybrid models namely Bagging based Reduced Error Pruning Trees (BREPT), MultiBoost based Reduced Error Pruning Trees (MBREPT), Rotation Forest-based Reduced Error Pruning Trees (RFREPT), Random Subspace-based Reduced Error Pruning Trees (RSREPT) for landslide susceptibility assessment and prediction. In total, ten topographical and geo-environmental landslide conditioning factors were considered for analyzing their spatial relationship with landslide occurrences. Receiver Operating Characteristic curve, Statistical Indexes, and Root Mean Square Error methods were used to validate performance of these models. Analysis of model results indicate that the BREPT is the best model for landslide susceptibility assessment in comparison to other models.