Typhoon-Driven Flood Occurrence Prediction in Western Visayas Using Machine Learning: A Comparative Study with SHAP-Based Feature Analysis

Western Visayas (Region VI) is among the Philippine regions most exposed to typhoon- and monsoon-driven flooding, yet it has not been addressed by region-specific, interpretable machine learning flood modeling. This study develops an explainable machine learning framework for typhoon-driven flood-risk prediction across the five Panay and Guimaras provinces of the region. Twenty-five years (2000 to 2024) of CHIRPS satellite daily rainfall and IBTrACS tropical-cyclone data were integrated into a dataset of 45,660 province-day records, with flood-risk days labeled using province-specific 90th-percentile rainfall thresholds. Four classifiers, namely Logistic Regression, Random Forest, XGBoost, and Support Vector Machine, were trained and compared under a time-aware train and test split that prevents temporal leakage, using years 2000 to 2020 for training and 2021 to 2024 for testing. All four models achieved high discrimination, with ROC-AUC values above 0.99. Random Forest provided the best overall balance of precision and recall, while XGBoost achieved perfect recall, missing no flood-risk days in the test period. SHAP analysis identified same-day and antecedent rainfall as the dominant predictors and revealed a sharp threshold near 27 to 32 millimeters per day at which predicted flood risk shifts decisively, providing an interpretable signal for disaster risk reduction and management. To the best of the author's knowledge, this is the first explainable, occurrence-level machine learning flood-prediction study for Western Visayas, and it establishes a reproducible baseline for further research in the region.