<![CDATA[Floods are the most common hydrometeorological disasters in Indonesia, and Bekasi Regency is among the areas with a high level of vulnerability. This study aims to develop a Twitter-based flood detection system using Support Vector Machines (SVM), with Logistic Regression (LR) and Random Forest (RF) as comparison models. A total of 4,436 tweets from the 2020–2024 period were collected using Tweet Harvest, manually labeled, and then processed through preprocessing, TF-IDF feature extraction, and data splitting using group-based random splitting (80:20). Evaluation was conducted using 10-fold GroupKFold cross-validation with recall as the primary metric, followed by hyperparameter tuning using GridSearchCV. The evaluation results showed that SVM performed best compared to LR and RF. The tuned SVM model achieved a test recall of 0.8673, exceeding the minimum threshold of 0.80. The model was then integrated into a web-based monitoring dashboard that displays interactive maps, statistics, and temporal trends. Black-box testing across nine scenarios achieved a 100% success rate, while a user experience evaluation using the UEQ-S on 18 respondents yielded an overall score of 1.97, categorized as “Very Good.” The research results indicate that the developed system is capable of effectively supporting social media-based flood monitoring.]]>
Copyrights © 2026
