Flood assessment using machine learning and its implications for coastal spatial planning in Phu Yen Province, Vietnam

Abstract

ABSTRACT The objective of this study was the development of a new machine learning model using a radial basis function neural network (RBFNN) to build flood susceptibility maps and damage assessment for the Phu Yen province of Vietnam. The built model will be optimized by five algorithms, namely Giant Trevally Optimization (GTO), Golden Jackal Optimization (GJO), Brown-Bear Optimization (BBO), Gray Wolf Optimizer (GWO), and Whale Optimization Algorithm (WOA) to find out the best model to establish the flood susceptibility map. These models were evaluated using the statistical indices such as root mean square error (RMSE), mean absolute error (MAE), receiver operating characteristic (ROC), area under the curve (AUC), and coefficient of determination (COD). The result showed that all five optimization algorithms were successfully improving the performance of the RBFNN model, among them the hybrid model RBFNN–BBO has the highest performance with AUC = 0.998 and R2 = 0.8 and the RBFNN–GTO model has the lowest performance with AUC = 0.755 and R2 = 0.65. The regions identified with a high- and very-high flood susceptibility area (1,075 km2) were concentrated on the plain and along three of the largest rivers in Phu Yen province.