Flood Risk Assessment Combining Machine Learning with Multi-criteria Decision Analysis in Jiangxi Province, China

Abstract Floods are one of the most frequent natural hazards worldwide. Accurate flood risk mapping is critical for effective flood management in flood-prone areas. In this study, we employed the multi-criteria decision analysis (MCDA) method to develop a flood risk map that combines flood susceptibility and vulnerability factors. Three machine learning models—random forest (RF), XGBoost, and LightGBM—were selected as the basic classifiers for creating flood susceptibility maps. Historical flood data and 13 flood-influencing factors were extracted for machine learning training. Model performance was assessed using precision, recall, accuracy, F1-score, and AUC through 5-fold cross-validation. All three models performed well, but RF slightly outperformed the other two according to the evaluation results. We used the analytic hierarchy process (AHP) method to combine the flood susceptibility map generated by the RF model with flood vulnerability indicators to produce a flood risk map. Our findings demonstrate that integrating advanced machine learning techniques with MCDA method offers an effective approach for flood risk assessment, providing a robust foundation for decision making in flood risk management.