Maize is a strategic commodity in Indonesia’s national food system, yet traditional yield-prediction methods based on statistical or survey approaches often fail to capture the nonlinear and dynamic relationships among agronomic, climatic, and socio-economic variables. Accurate forecasting remains essential for supporting food self-sufficiency and climate-resilient agricultural planning. To address these challenges, this study proposes SMART-JAGUNG, a machine learning–based maize yield prediction system employing three ensemble and regression models: Random Forest (RF), Support Vector Regression (SVR), and eXtreme Gradient Boosting (XGBoost). The dataset comprises five years of maize production data from the Indonesian Central Bureau of Statistics (BPS), along with auxiliary variables including rainfall, temperature, NDVI, seed type, and fertilizer use. Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination before and after hyperparameter tuning with GridSearchCV. Results indicate that RF achieved the best performance before tuning (MAE = 36,310.53; RMSE = 95,343.05; = 0.9758), followed closely by XGBoost, while SVR consistently underperformed. Although post-tuning performance slightly decreased, the predicted-versus-actual visualization confirmed the robustness of RF and XGBoost for non-extreme data. Overall, SMART-JAGUNG demonstrates strong potential as a reliable, data-driven decision-support tool for precise maize yield estimation, contributing to sustainable food security and national self-sufficiency policies.
Lubis et al. (Wed,) studied this question.