A Novel Machine Learning Approach for Flood Prediction with Local Interpretable Explanations

Given the devastating impact of floods, pinpointing flood-prone areas is vital for improving disaster management. While current data-driven methods have achieved high accuracy in flood prediction, they often lack interpretability. This paper addresses this gap by developing a two-part machine learning system that first identifies flood-prone regions and then deter-mines key contributing features. Using a Random Forest model, we achieved 91% accuracy. We then employ LIME to interpret the predictions, revealing that elevation, distance to rivers, slope, and geology significantly impact a region's flood susceptibility. This research marks a potential advancement toward more comprehensive flood predictions, enabling targeted mitigation strategies.