Flood risk assessment based on machine learning and multi-criteria decision analysis

Abstract In this study, three machine learning algorithms, including Multilayer perceptron (MLP), Radial basis function (RBF) and Back propagation (BP), to analyse the flood susceptibility of the study area. Considering the flood susceptibility, hazard and exposure of the study area, the Multi-Criteria Decision Analysis (MCDA) was used to conduct a flood risk assessment of the study area. Collection of information on the physical geography and socio-economic distribution of the study area, including Rainfall, DEM, Slope, Land use/Land cover (LULC), Distance from the rivers, Curve number (CN), Manning coefficient, topographic wetness index (TWI), population density and Gross Domestic Product (GDP) distribution. A machine learning approach to assess flood susceptibility in the study area, using six standard statistical indices (accuracy, precision, recall, F-score, Root Mean Square Error and Kappa index) and Receiver Operating Characteristic (ROC) curves called AUG to evaluate the performance of the machine learning model. The results show that all the proposed models perform well, but performance of the MLP model is the best (AUC = 0.951). Using the Analytic Hierarchy Process (AHP) method to determine the weights, the flood susceptibility map obtained from the MLP is overlaid with hazard and vulnerability for flood risk assessment. The results show that integrated machine learning algorithms and MCDA for flood analysis are a promising approach to flood risk assessment.