Peningkatan Akurasi Prediksi Curah Hujan menggunakan Gradient Boosting dan CatBoost dengan Pendekatan Voting Classifier

Accurate rainfall prediction is essential for agriculture, disaster mitigation, and water resource management, especially in the face of climate change impacts. This research aims to improve the accuracy of rainfall prediction using gradient boosting and CatBoost with a voting classifier approach. The data used in this study amounted to 1,461 based on weather data from BMKG Semarang City (2020-2023). The data was analyzed using the Gradient Boosting and CatBoost algorithms with a voting classifier framework. The input features include temperature (Tn, Tx, Tavg), humidity (RH_avg), rainfall (RR), length of irradiation (ss), wind speed (ff_x, ff_avg), and wind direction (ddd_x). The GridSearchCV technique was used for hyperparameter optimization. The model predicts based on rainfall intensity categories, namely no rain, light rain, moderate rain, heavy rain, and extreme rain. The results showed that the model with optimization and ensemble approach achieved 87.89% accuracy, 0.88 precision, 0.88 recall, 0.88 f1-score, and 0.8486 cohen's kappa. Meanwhile, gradient boosting and CatBoost individually produced 75.99% and 85.68% accuracy. With these data input features, the model is able to predict extreme rainfall categories that match the actual data. This research is an important contribution to the development of early weather warning systems, disaster mitigation, and climate management.