A Comparative Study of Random Forest and XGBoost Algorithms for Oil Palm Yield Prediction and Sustainable Land Use Classification

The oil palm industry has experienced substantial growth as one of the world’s leading vegetable oil producers. As the global palm oil industry continues to expand, data-driven approaches are increasingly needed to predict land productivity and optimize land-use management. This study compares the performance of the Random Forest and XGBoost nonlinear regression algorithms in predicting oil palm plantation productivity, identifies the most influential variables affecting productivity, and provides field-level recommendations for classifying plantation land suitability for bioenergy development or continued oil palm production. The dataset consisted of 128 historical plantation observations, including variables such as plant age, tree density per hectare, fertilization percentage, rainfall, and seed type. Model performance was evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and the coefficient of determination (R²). The results showed that XGBoost outperformed Random Forest, achieving an R² value of 0.883, MSE of 7.38, and RMSE of 2.72, compared to Random Forest with an R² value of 0.871, MSE of 8.05, and RMSE of 2.84. Feature importance analysis and correlation heatmap results indicated that Trees per Hectare and Plant Age were the most influential variables, exhibiting the strongest relationships with plantation productivity. Based on the productivity threshold of 13.9 tons/ha, plantation blocks with productivity below this threshold are recommended for consideration as potential bioenergy plantation areas. This study demonstrates that the XGBoost algorithm can support effective decision-making in sustainable land-use planning within the oil palm industry.