Machine Learning Ensemble Model for Flood Susceptibility Mapping

Floods can be considered the most dangerous natural disaster, given their unpredictability and capacity to wipe out valuable life and property. The timely and efficient prediction of floods and flood susceptible or risk zones has been of utmost importance and can help with risk assessment, long-term management, and future preparedness. Over the years, Machine Learning (ML) has evolved as a powerful tool to build accurate flood models, inundation maps, and warning systems. This study uses the Support Vector Machine (SVM) and Random Forest (RF) algorithms and a Bagging Ensemble Model for Flood Susceptibility Mapping of Ernakulam, Kerala, India using multi-source geospatial data and the WEKA software. A total of twelve Flood Conditioning Factors (FCFs) are considered as variables, namely elevation, slope, curvature, Topographic Roughness Index (TRI), Topographic Wetness Index (TWI), Stream Power Index (SPI), rainfall, Land Use Land Cover (LULC), distance to the river, drainage density, and geology. The contribution of factors is assessed using the OneR Feature Selection method. The models are compared using the Receiver Operating Characteristics (ROC) curve and the Area Under the Curve (AUC) method. The Flood Susceptibility Map provides an opportunity for planners and authorities to flood preparation and long-term planning against flood impacts.