Forecasting Malaysian Palm Oil Futures Using Random Forest and LSTM Networks: Enhanced Predictive Modeling with Climate-Based Parameters

This study presents a comprehensive analysis of machine learning approaches for predicting palm oil futures prices in Malaysia. The research employs Random Forest (RF) and Long Short-Term Memory (LSTM) models to evaluate the impact of both conventional market indicators and climate factors on price prediction accuracy. Through rigorous analysis of 112 potential indicators across agricultural, commodity, Malaysia-specific, economic, and climate categories, 83 core indicators were identified for the predictive models. Results demonstrate that the RF model consistently outperformed LSTM, achieving a Mean Absolute Percentage Error (MAPE) of 7.45% compared to LSTM's 13.36% in multivariate analysis. The incorporation of climate factors significantly improved LSTM model performance, reducing its MAPE from 21.59% to 13.36%, while showing minimal impact on RF model accuracy. Notably, the RF model exhibited superior performance in long-term (100-day) forecasting compared to short-term (15-day) predictions, with MAPE values of 3.61% and 7.45% respectively. This research provides valuable insights into agricultural commodity trading and risk management, particularly highlighting the effectiveness of RF models for palm oil futures price prediction and the potential benefits of incorporating climate variables in forecasting models.