Integration of AHP and machine learning methods for flood susceptibility analysis in a Meandering River

Floods, caused by the overflow of water from natural channels, are among the most destructive natural hazards, affecting human life, property, and ecosystems. Their impact is increasingly significant due to climate change and human-induced land use changes. This study aims to evaluate flood susceptibility in the Sarayköy district of Denizli province using spatial approaches and to compare the predictive performance of different modeling techniques. Four models were applied: Analytic Hierarchy Process (AHP), Maximum Entropy (MaxEnt), Random Forest (RF), and Support Vector Machines (SVM). While AHP relies on expert judgment and hierarchical weighting of criteria, MaxEnt, RF, and SVM are machine learning-based approaches. Model performance was assessed using the Area Under the Receiver Operating Characteristic Curve (AUC-ROC). Results show that MaxEnt achieved the highest accuracy (AUC = 0.86), followed by RF (0.82), SVM (0.79), and AHP (0.73), highlighting the superior predictive capability of machine learning methods compared to traditional techniques. Machine learning models demonstrated particularly high accuracy in river channels and low-gradient plains, indicating their applicability for disaster risk management. Although AHP produced broader and less sensitive classifications, it remains valuable for rapid preliminary assessments, especially in data-scarce regions. Overall, this study confirms that numerical and spatial analysis of flood risk can be effectively conducted using machine learning approaches, and future research should explore model application across diverse regions, integration of additional hydro-meteorological parameters, and combined modeling strategies to improve risk prediction. Such advances will support more effective, rapid, and spatially-informed decision-making in flood risk management.