Detection of areas prone to flood-induced landslides risk using certainty factor and its hybridization with FAHP, XGBoost and deep learning neural network
Romulus Costache, Sk Ajim Ali, Farhana Parvin, Quoc Bao Pham, Alireza Arabameri, Hoang Nguyen, Anca Crăciun, Duong Tran Anh
Abstract
This article is intended to assess the flood-induced landslide susceptibility in the Indian state of Assam. This study area has high frequency and severity of landslides that are triggered by heavy rainfall and floods. In order to obtain the results, two machine learning models (XGBoost and DLNN) and one of the fuzzy-multi-criteria decision-making methods (FAHP) were combined with certainty factor (CF) bivariate statistic model. Firstly, 16 landslide predictors and 198 flood-induced landslide locations were prepared, this data set being split into training (70%) and validating data sets (30%). The analysis of the results shows that the region’s most prone to flood-induced landslide occurrence can be found in the southern part, while those less prone to these phenomena are generally located in the northern part of the study area. Receiver Operating Characteristic (ROC) curve indicator shows that XGBoost-CF is the most performance model (area under the curve [AUC] = 0.977), followed by FAHP-CF (AUC = 0.976), DLNN-CF (AUC= 0.974) and CF (AUC = 0.963).