Geospatial modeling using hybrid machine learning approach for flood susceptibility

Advanced methods for flood susceptibility mapping are required to minimize hazards in the watershed. Here, Partial Least Square-Structural Equation Model (PLS-SEM) was introduced to analyze the impact of flood influencing factors. PLS-SEM integrated with four Machine Learning (ML) methods as Multi-Layer Perceptron Neural Network (MLPNN), K Nearest Neighbor (KNN), Support Vector Machine (SVM) and Radial Basis Function Neural network (RBFN). In addition, significant flood influencing factors from PLS-SEM analysis was taken as the input of ML models. Then SVM, MLPNN, KNN, and RBFN integrated with the PLS-SEM classifier to develop hybrid models for constructing FSM. The performance of models is assessed in terms of standard statistical methods. The performance of the achieved model is good having AUROC > 0.8 and PLS-SEM-SVM (AUROC = 0.978) perform superior than others. Thus, hybrid SVM model can be best utilized for flood susceptibility. This study provides the importance of mechanism for flood influencing factors and extends the application of proposed hybrid ML models to minimize flood risk.