Comparative Analysis of Machine Learning Algorithms for Weather-Based Oil Palm Production Prediction

Oil palm production is strongly influenced by climatic variability, making accurate production prediction important for plantation management and decision-making. However, most previous studies have focused on specific machine learning algorithms without explicitly considering the delayed physiological response of oil palm plants to weather conditions.. This study proposes a comparative machine learning framework for oil palm production prediction using a 24-month lag representation based on climatic variables, including rainfall, rainy days, temperature, humidity, wind speed, and sunshine duration. Several machine learning algorithms were evaluated, namely Random Forest, XGBoost, ANN, LightGBM, SVR, and MLR. Hyperparameter tuning was performed using Grid Search with 5-fold cross-validation. Experimental results show that SVR and LightGBM achieved the best predictive performance with an R² value of 0.75. SVR produced the lowest MAPE of 10.92%, while LightGBM achieved the lowest RMSE of 515.69 tons. Feature importance analysis indicates that sunshine duration and rainy days contributed significantly to oil palm production prediction. Overall, the results demonstrate that kernel-based and boosting-based approaches are more effective in modeling nonlinear relationships between climatic variables and oil palm production after applying a 24-month lag representation.