Forecasting Natural Rubber Price in Malaysia using ARIMA and Long Short-Term Memory

Abstract

Natural rubber is a critical agricultural commodity in Malaysia, contributing significantly to the national economy and the global rubber market. However, the pricing of Standard Malaysia Rubber 20 (SMR20) is highly volatile, influenced by diverse economic and environmental factors, posing challenges for stakeholders. The objectives of this study are to apply reliable forecasting models for SMR20 prices in Malaysia, assess their accuracy using performance metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE), and identify the best model that provides the most accurate predictions for decision-making and market stability. The study finds that the Long Short-Term Memory (LSTM) model outperforms the ARIMA model in forecasting SMR20 prices, demonstrated by its superior accuracy and lower error metrics: MAE of 791.706, RMSE of 88.383, and MAPE of 9.398%. While ARIMA, based on the Box-Jenkins methodology, provides a reasonable fit for traditional time series data, it struggles to capture nonlinear dependencies in price patterns, whereas LSTM excels at modelling complex, non-linear relationships and long-term trends. These findings highlight the potential of advanced machine learning techniques like LSTM in agricultural commodity forecasting. The study emphasizes the significance of machine learning models in forecasting systems to enhance decision-making, risk management, and market stability in Malaysia's rubber industry. Future research should explore incorporating external variables, such as climate dynamics and global economic factors, to further improve forecasting accuracy and expand model applicability.