Comparative machine learning for flood susceptibility in Subarnarekha Basin, Odisha

Floods present significant threats to ecological balance, infrastructure, and socioeconomic stability, underscoring the necessity of accurate flood susceptibility modeling for effective disaster preparedness and mitigation. This study evaluates the performance of five machine learning (ML) algorithms—Logistic Regression (LR), K-Nearest Neighbours (KNN), AdaBoost, XGBoost, and Artificial Neural Networks (ANN)—for flood susceptibility prediction in the Subarnarekha River Basin, Odisha. Using a 30-year dataset comprising meteorological and hydrological variables from the monsoon season, model performance was assessed using established metrics such as Area Under the Receiver Operating Characteristic Curve (AUC-ROC). The results indicate that ANN achieved the highest AUC-ROC (0.87), demonstrating superior classification capability, while Logistic Regression (0.86) and XGBoost (0.85) also exhibited strong predictive performance. The findings highlight the importance of selecting appropriate ML models for flood risk assessment and suggest that integrating additional environmental factors and optimizing hyperparameters could further enhance prediction accuracy.