FLOOD SUSCEPTIBILITY MAPPING USING MACHINE LEARNING

Abstract

This study presents an innovative approach to flood susceptibility mapping in the Sylhet district of Bangladesh, integrating Geographic Information System (GIS), machine learning techniques, and the Analytic Hierarchy Process (AHP). The methodology involves the use of Digital Elevation Model (DEM) data, Landsat imagery, and soil data, complemented by a Python-based rainfall prediction model. The study employs AHP to weigh various floodcausing criteria and incorporates machine learning for predictive accuracy in rainfall forecasting. Two separate models are developed using random forest regressor for Sylhet sadar and kanaighat and the model had a R square value of .6677 and .5953 respectively with normalized root square mean error of 0.07 and 0.096. The Flood susceptibility map with prediction model shows an improvement in predicting vulnerable flood zones in comparison to the flood susceptibility map using historical rainfall data.