Analisis Perbandingan Model XGBoost dan Random Forest untuk Prediksi Potensi Banjir di DKI Jakarta

Flooding in DKI Jakarta is a hydrometeorological disaster that requires an accurate early warning system. The main problem with current prediction systems is the limitation in integrating climate data from upstream (Bogor) with hydrological conditions downstream (Jakarta). This research aims to build and compare machine learning classification models to address this issue. The methods used are Extreme Gradient Boosting (XGBoost) and Random Forest, utilizing integrated data covering rainfall (Jakarta and Bogor) and Water Level (TMA) at the Katulampa and Manggarai floodgates for the period of January 2024 to March 2025. The models were evaluated using hyperparameter tuning (GridSearchCV, RandomizedSearchCV) on two data split schemes (70:30 and 80:20). The research results show that the XGBoost model with GridSearchCV on the 80/20 scheme provided the best performance with an accuracy of 82.61%. The best Random Forest model (GridSearchCV, 80/20) achieved an accuracy of 80.43%. Although XGBoost was numerically superior, the results of the McNemar's Test indicated that there was no statistically significant difference (calculated x2 0.25 < table x2 3.841) between the performance of the two models.