<![CDATA[This Solar energy has become a strategic solution to meet energy demand on the island of Java, but its utilization is hindered by the variability of solar radiation. This research aims to develop an accurate long-term prediction model for solar radiation using an ensemble learning approach. The data used are historical data from 1984 to 2024 from NASA POWER. Three approaches are applied, namely Bagging with Decision Tree, Boosting using XGBoost, and a weighted average combined model. The models are evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R² Score. The research results show that the Bagging method produces an average MSE of 0.148, RMSE of 0.379, and R² of 0.574. The XGBoost method is superior with an average MSE of 0.140, RMSE of 0.368, and R² of 0.596. Meanwhile, the combined model with a weight of 0.9 on XGBoost and 0.1 on Bagging provides the best performance with an average MSE of 0.140, RMSE of 0.367, and R² of 0.598. This result proves that the combination of models with the right weighting can improve prediction accuracy, even though the difference is slight, and produce more stable predictions. This finding can support the planning and development of more efficient and sustainable solar energy infrastructure on the island of Java. Keywords – Prediction; Ensemble Learning; Solar Radiation; Java Island; Renewable Energy.]]>
Copyrights © 2026
