Oil Palm Production Prediction Using RNN-LSTM

Authors: Muhdan Syarovy | Year: 2025 | DOI: https://doi.org/10.17605/OSF.IO/3KX5Z

---

Accurate oil palm production forecasting is key to plantation operational efficiency — from determining harvesting schedules to managing transportation logistics and supply chains. As deep learning technology advances, more accurate methods are emerging, one of which is Recurrent Neural Networks (RNN) combined with Long Short-Term Memory (LSTM). This method is specifically designed to analyze time-series data, making it a natural fit for oil palm production patterns that inherently vary over time.

This study trained an RNN-LSTM model on historical FFB production data spanning several years from a specific plantation. The model architecture was designed to capture long-term dependencies in the data — the key differentiator of LSTM compared to standard RNN which is prone to forgetting early patterns when dealing with long data sequences. Hyperparameter optimization was performed to achieve the best predictive performance before the final model was evaluated on test data not used during training.

The results demonstrated that RNN-LSTM is capable of predicting oil palm production with a high degree of accuracy, with prediction errors significantly lower than conventional statistical forecasting methods. The model proved particularly effective at capturing seasonal patterns and production trends — information highly valuable for medium-term production planning at the plantation level.

The following table summarizes the key aspects of the study.

Aspect	Role in Study	Method / Approach	Key Advantage	Main Finding
RNN-LSTM Model	Deep learning architecture designed for time-series forecasting	LSTM gates control long-term memory, preventing gradient vanishing in long sequences	Very High	Significantly lower prediction errors compared to conventional statistical forecasting methods (ARIMA, Holt-Winters)
Data Requirements	Historical FFB production data for model training and validation	Multiple years of monthly data used, split into training and test sets	Significant	Model quality improves with longer historical data. Minimum 3 years recommended for acceptable accuracy
Seasonal Pattern Capture	Highly accurate identification of cyclical production fluctuations	LSTM memory cells retain information across seasonal cycles	Significant	Excellent at modeling seasonal peaks and troughs typical of oil palm production curves
Hyperparameter Optimization	Fine-tuning model parameters for optimal performance	Grid search over learning rate, number of layers, number of units, dropout rate	Limited	Tuning significantly improves performance. Optimal architecture: 2 LSTM layers with 64 units each, dropout 0.2
Practical Deployment	Implementation in plantation operational decision-making	Standalone Python script or web-based dashboard with periodic retraining	Significant	Can be deployed with moderate Python expertise. Periodic retraining needed monthly/quarterly for sustained accuracy

This research suggests that deep learning methods like RNN-LSTM can be an effective tool to support production planning in oil palm plantations, though successful implementation requires adequate data availability and in-house technical capability to develop and maintain the model.

Have questions about implementing RNN-LSTM in your plantation? Write them in the comments below!

---