Enhancing Flood Severity Prediction through a Combined Threshold-Based Alert Algorithm and Random Forest Classifier

Flooding is a significant and recurrent issue that poses severe risks to communities, economies, and environments. Traditional flood prediction methods, which often rely on fixed thresholds, may lack adaptability to changing conditions. Meanwhile, machine learning (ML) models like Random Forest Classifiers (RFCs) offer advanced predictive capabilities, but may not provide straightforward alert mechanisms. This research proposes a novel approach that integrates a Threshold-Based Alert Algorithm (TBA) with an RFC to enhance flood severity prediction. The combined model leverages the strengths of both methodologies: immediate, interpretable alerts from the TBA and adaptive, data-driven insights from the RFC. Using historical and real-time data from river basins in Wisconsin and New York, the hybrid model demonstrates improved accuracy and reliability over traditional methods and standalone RFCs. The results indicate that this integrated approach offers a more effective tool for flood management and disaster mitigation.