The using of deep learning hybrid model and meta-heuristic algorithms in flood susceptibility assessment

In recent years, instances of flood events have unfortunately increased significantly in both magnitude and frequency due to climate change, population growth, and accelerating urbanization. This study developed Flood Susceptibility Maps (FSMs) for Qom Province, Iran, using a database comprising twelve conditioning factors derived from Remote Sensing (RS) datasets and Geographic Information System (GIS). The factors were weighted using two Multi-Criteria Decision-Making (MCDM) techniques, i.e., Fuzzy DEMATEL and the Best–Worst Method (BWM). After weighting the input variables, the pairing of Deep Belief Network (DBN) and Support Vector Machine (SVM) classifier was examined. The resulting model was further optimized using metaheuristic algorithms. Hybrid models (DBN-SVM), (DBN-SVM-PSO), (DBN-SVM-HHO), and (DBN-SVM – (PSO-HHO)) were evaluated using the Area Under Curve (AUC) method along with other statistical indicators. Based on the evaluation results, the DBN-SVM-(PSO-HHO) model achieved relatively better performance among the evaluated models. The proposed hybrid approach helped the identification of flood-susceptible zones. The results of the hybrid model can support urban managers and decision-makers in improving urban planning and management.