An Application of Hybrid Bagging-Boosting Decision Trees Ensemble Model for Riverine Flood Susceptibility Mapping and Regional Risk Delineation
Javeria Sarwar,Saud Ahmed Khan,Muhammad Azmat,Faridoon Khan
DOI: https://doi.org/10.1007/s11269-024-03995-6
IF: 4.426
2024-10-10
Water Resources Management
Abstract:Flood disasters have become the most prevalent natural phenomenon as a result of climate change and other environmental factors. Most countries are vulnerable to flooding, which is a serious hazard to human life worldwide and has negative effects on the physical, social, and economic spheres. The utilization of ensemble machine learning algorithms has experienced a significant increase in the field of machine learning due to their resilience and ability to handle data that contains noise. These algorithms enhance the precision of forecasts by blending the results from multiple feeble decision models. Therefore, the research is an endeavor to propose a novel hybrid bag-boost decision tree ensemble model for riverine flood susceptibility mapping. The ensemble model integrates four independent decision tree models namely, Random Forest (RF), Logistic Model Tree (LMT), Naïve Bayes Tree (NBT), and Reduced Error Pruning Tree (REPT). For susceptibility mapping, a spatial database is constructed by considering 5500 flood spots and an equivalent number of non-flood points. The flood conditioning factors considered for the research possess environmental, topographic and human induced factors. The dataset has been randomly segregated into sample sizes of 70% and 30% for training and validating the models, respectively. The performance of the proposed ensemble model is assessed by utilizing various statistical evaluation measures; accuracy, precision, Receiver Operating Characteristic (ROC) curve, Friedman test and Neymenyi test, and is compared with the stand-alone decision tree models. The performance outputs of the models revealed that the hybrid bag-boost decision tree ensemble model (RF-LMT-NBT-REPT) performed the best with a 99.5% accuracy level for the training sample and 98.9% for the validating sample. The inundation maps are hence acquired by utilizing the hybrid bag-boost ensemble model for the years 2022 and the predicted flood of 2032 and regional hazard analysis has been performed. The study proposes that the hybrid bag-boost decision tree ensemble model be utilized for an accurate and precise hydraulic modelling and susceptibility analysis.
water resources,engineering, civil