Enhancing Flood Susceptibility Modeling: a Hybrid Deep Neural Network with Statistical Learning Algorithms for Predicting Flood Prone Areas

Flooding, with its environmental impact, represents a naturally destructive process that typically results in severe damage. Consequently, accurately identifying flood-prone areas using state-of-the-art tools capable of providing precise estimations is crucial to mitigate this damage. In this study, the objective was to assess flood susceptibility in Lorestan, Iran, through the utilization of a novel hybrid approach that incorporates a Deep Neural Network (DNN), Frequency Ratio (FR), and Stepwise Weight Assessment Ratio Analysis (SWARA). To achieve this, a geospatial database of floods, comprising 142 flood locations and 10 variables influencing floods, was employed to predict areas susceptible to flooding. Frequency Ratio (FR) and Stepwise Weight Assessment Ratio Analysis (SWARA) were utilized to assess and score the variables influencing floods. Simultaneously, DNN, recognized as an excellent tool in machine learning and artificial intelligence, was employed to generate the inferred flood pattern. The models' performance was evaluated using metrics such as the area under the curve (AUC), receiver operating characteristic (ROC) curve, and various statistical tests. The results indicated that both proposed algorithms, DNN-FR and DNN-SWARA, were able to estimate future flood zones with a precision exceeding 90%. Furthermore, the outputs confirmed that, although both algorithms demonstrated high goodness-of-fit and prediction accuracy, the DNN-FR (AUC = 0.953) outperformed the DNN-SWARA (AUC = 0.941). Consequently, the application of the DNN-FR algorithm is proposed as a more reliable and accurate tool for spatially estimating flood zones.