Analysis of the Comparison Between Linear Regression, Random Forest, and Logistic Regression Methods in Predicting Crude Palm Oil (CPO) Price

Abstract

This study compared the predictive capabilities of linear regression, random forest, and logistic regression models for forecasting crude palm oil prices. Utilizing historical data from 02/11/2020 to 13/11/2023, the dataset underwent training and testing with three scenarios: 90:10, 80:20, and 70:30. Evaluation metrics, including RMSE, MSE, and MAPE, assessed model performance. Each method had unique strengths and weaknesses, and the choice depended on application needs. The goal was to improve decision accuracy in predicting crude palm oil price trends. In the 90:10 scenario, random forest outperformed linear and logistic regression, yielding smaller MSE (43948.56), MAE (80.37), and RMSE (209.64). Similarly, in the 80:20 scenario, random forest had smaller MSE (137787.61), MAE (106.38), and RMSE (371.20). In the 70:30 scenario, random forest showed smaller MSE (107582.32), MAE (104.13), and RMSE (328). Overall, random forest consistently demonstrated better performance than linear and logistic regression.