Averaged Neural Network Integrated with Recursive Feature Elimination for Flood Hazard Assessment

Abstract

This article proposes a novel method for identifying flooded areas with high accuracy using information from hydro-environmental features and Radar images. A combination of averaged neural networks (avNNet) and feature extraction algorithms were used to achieve this goal. The recursive feature elimination (RFE) method was utilized to Figure out the relevant features. Then, the avNNet was employed on these features to classify/identify hazardous areas. Based on the outcomes of the RFE method, six variables of distance from river, elevation, vegetation, drainage density, precipitation, and slope were the most crucial influencing variables for flood hazard modeling in the area. In a nutshell, according to the results, the avNNet model achieved an accuracy of more than 96% and Kappa values greater than 93% for different used return periods.