<![CDATA[Fresh Fruit Bunch (FFB) productivity in oil palm exhibits seasonal patterns that pose challenges for predictive modeling, particularly given the limited amount of data. This study aims to compare the performance of Neural Networks and Random Forests in predicting FFB productivity based on temporal features, including lag, rolling mean, and cyclical encoding. Evaluation was conducted using time-series validation with MAE, RMSE, and R² metrics. The results indicate that Neural Networks face generalization limitations with limited data, reflected in poor performance on the test data. Conversely, Random Forest delivers more stable and accurate performance with an MAE of 0.2581, an RMSE of 0.3325, and an R² of 0.9675. These findings confirm the superiority of tree-based ensemble approaches in handling seasonal data with small sample sizes. The contribution of this research is to provide empirical evidence and recommendations for more reliable models for TBS productivity prediction as a basis for developing decision support systems in the plantation sector.]]>
Copyrights © 2026
