<![CDATA[Gorontalo Province regularly faces flood hazards driven by a combination of high rainfall, low-lying topography, and limited river capacity, while conventional hazard mapping approaches based on scoring and overlay remain limited in capturing the non-linear relationships among flood-causing variables. This study aims to produce a more accurate flood susceptibility map for Gorontalo Province by integrating climate and spatial data using machine learning. Seven physical variables (topography, slope, rainfall, land cover, soil type, DEMNAS elevation, and river distance) were paired with 186 historical flood points from 2024–2025 and non-flood control points, then modeled using five algorithms—Logistic Regression, Random Forest, Support Vector Machine, Multi-Layer Perceptron, and XGBoost—compared through 5-fold stratified cross-validation. The best-performing model was then used to predict the flood susceptibility index across the entire study grid and classify it into five hazard classes using the Natural Breaks (Jenks) method. The results show that Random Forest achieved the best performance, with a mean AUC of 0.9792 and a mean accuracy of 95.68%. Independent validation against historical flood points showed that 179 of 186 points (96.2%) fell within areas classified as hazardous, confirming the spatial validity of the resulting map. This study concludes that integrating climate and spatial data through machine learning produces more accurate flood susceptibility mapping than conventional approaches, and can serve as a decision-support tool for local government disaster mitigation policy. ]]>
Copyrights © 2026
