<![CDATA[West Java Province exhibits high vulnerability to hydrometeorological disasters due to dynamic rainfall variability, necessitating an accurate weather prediction system for effective disaster mitigation.1 This study aims to conduct a comparative performance analysis of Machine Learning algorithms, specifically Support Vector Machine (SVM), Naïve Bayes, Random Forest, and XGBoost, in predicting rainfall events based on 2024 daily meteorological data sourced from BMKG. Through computational experiments utilizing three data splitting scenarios 80:20, 75:25, and 70:30, and Recursive Feature Elimination (RFE), the results demonstrate that Naïve Bayes, Random Forest, and XGBoost consistently achieved a perfect accuracy of 100% across all scenarios, whereas SVM exhibited stable but more conservative performance with an average accuracy of 95.4%. In-depth analysis indicates that the absolute accuracy achieved under specific data conditions was significantly influenced by the dominance of the daily rainfall feature (RR), leading to indications of data leakage where ensemble and probabilistic models exploited deterministic relationships much more effectively than SVM. Consequently, this study recommends a rigorous re-evaluation of input features, prioritizing atmospheric leading indicators, to develop a more realistic and adaptive early warning system in the future.]]>
Copyrights © 2026
