Predicting Palm Oil Price Direction using Random Forest

Abstract

The palm oil market in Myanmar greatly depends on the world palm oil price changes, especially on the price changes in the export countries of palm oil to Myanmar. As palm oil market in Myanmar represents the back bone of Myanmar Edible Oil Dealers Association (MEODA), we propose the predictive model to aid in decision making process of palm oil importers whether they should conduct import transaction today or not in this paper. This prediction of palm oil price condition in Myanmar has been taken on the previous dataset supported by MEODA to eliminate ever-increasing risks and uncertainties in the future. This model will forecast whether the price of palm oil in Myanmar will rise or not in 14 days from today, the length of period is necessary to be ready to trade imported palm oil in local market for Myanmar importers. Our model is trained using C4.5 Random Forest Classification Algorithm on the palm oil market dataset from MOEDA. Hyperparameter tuning techniques are conducted to analyze whether the predictive performance can be enhanced. From the obtainable dataset in Myanmar palm oil market, the predictive model with chosen hyperparameters set achieves the prediction accuracy of 91.11% on the test dataset.