Predicting and Mapping Flood Susceptibility: Leveraging Explainable AI and GIS Techniques

Flood susceptibility mapping is crucial for disaster management, but traditional methods struggle with precision and adaptability in the face of climate variability. This study employed a Random Forest Classifier (RFC) to enhance prediction accuracy in flood susceptibility mapping. Exploratory data analysis (EDA) was conducted to examine data characteristics and identify key flood-influencing variables. Feature engineering techniques were used to refine predictors and determine an optimal subset of features. The model exhibited high predictive power with an ROC-AUC value of 0.92. To address the challenge of communicating complex results, plots of partial dependence and permutation feature importance were generated. The findings were translated into actionable flood susceptibility maps using GIS software. This approach bridges the gap between advanced machine learning techniques and practical insights, enabling policy-makers to visualize flood-prone areas and implement targeted mitigation strategies.