Improving Flood Susceptibility Modeling Through a Multi-Criteria Feature Selection Framework

Flood susceptibility mapping is essential for effective flood mitigation and land management, particularly in rapidly urbanizing watersheds. This study proposes a multi-criteria framework for improving flood susceptibility modeling through conditioning factor selection. Nineteen topographic, hydrological, environmental, and land-use factors were used to develop four models: Frequency Ratio (FR), Shannon Entropy (SE), Certainty Factor (CF), and VIKOR. A flood inventory of 230 historical flood locations was compiled, with 160 locations (70%) used for model development and 70 (30%) reserved for independent validation. Model performance was evaluated using Receiver Operating Characteristic (ROC) analysis based on Success Rate and Prediction Rate datasets. A factor selection framework integrating ROC-AUC, Information Value (IV), Pearson correlation, and Variance Inflation Factor (VIF) was then applied to identify informative and non-redundant variables. The framework reduced the conditioning factors from 19 to 9. All optimized models showed improved predictive performance. The FR model showed the greatest improvement, with Success Rate AUC increasing from 58.45% to 90.41% and Prediction Rate AUC from 58.10% to 89.88%. The CF model achieved the highest overall performance after optimization, while SE and VIKOR also improved substantially. The proposed framework improves prediction while reducing input-data complexity.