Enhancing flood prediction in Southern West Bengal, India using ensemble machine learning models optimized with symbiotic organisms search algorithm

Abstract

In regions with limited flow and catchment data needed for the configuration and calibration of hydraulic and hydrological models, employing spatial flood modeling and mapping enables authorities to predict the spatial extent and severity of floods. This study leveraged flood inventory data, coupled with various conditional variables, to formulate a novel Ensemble model. This ensemble model combined four hybridized models based on Support Vector Machine (SVM), Naïve Bayes (NB), Decision Classification Tree (DCT), and Artificial Neural Network (ANN), all of which were optimized using the metaheuristic Symbiotic Organisms Search algorithm (SOS). The precision of the flood inundation map generated by the four hybrid models and the ensemble model was assessed using standard metrics. The results demonstrated that the ensemble model outperformed other models, with an accuracy metric of 0.99 Area Under the Curve (AUC) during the training stage and 0.96 during the testing stage. This underscores the effectiveness of the ensemble approach in flood preparedness and response applications. Furthermore, a comparison was conducted, comparing the performance of the developed ensemble model against other studies within the state of West Bengal. The findings highlighted a significant improvement in the ensemble model's performance with an AUC score of 0.96 in validation compared to studies in similar areas within West Bengal with AUC score ranged from 0.73 to 0.92. In conclusion, the methodology employed in this study holds promise for application in other regions worldwide that face challenges related to limited data availability for accurate flood inundation mapping.