A Novel Hybrid Intelligent Approach of Random Subspace Ensemble and Reduced Error Pruning Trees for Landslide Susceptibility Modeling: A Case Study at Mu Cang Chai District, Yen Bai Province, Viet Nam

Abstract

In the present study, a hybrid approach of Random Subspace Ensemble (RSS) and Reduced Error Pruning Trees (REPT) has been proposed to create a novel hybrid model namely RSS-REPT for landslide susceptibility modeling of the Mu Cang Chai district, Yen Bai province of Vietnam where is affected by a number of landslides every year. For the development of model, a spatial database consisting of 248 historic landslide events and 15 affecting factors (slope, aspect, curvature, plan curvature, profile curvature, elevation, lithology, land use, distance to faults, fault density, distance to roads, road density, distance to rivers, river density, and rainfall), was constructed to generate training and testing datasets. The novel hybrid model was then constructed using training dataset for landslide susceptibility assessment, and its predictive capability was validated using Receiver Operating Characteristic (ROC) curve and Statistical Indexes (SI) analysis. Performance of this novel model has been compared with another popular model namely Support Vector Machines (SVM). Results indicate that its performance (AUC = 0.835) is higher in comparison to the SVM model (AUC = 0.804). Thus the RSS-REPT can be considered as one of the promising methods for better landslide susceptibility assessment of landslide prone areas.