<![CDATA[The prediction of palm oil prices plays a strategic role in decision-making within the agribusiness sector, particularly in addressing market volatility and imbalanced historical data distribution. This study aims to optimize the accuracy of palm oil price prediction by applying a stacking approach that combines machine learning and deep learning algorithms, while integrating the Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance issues. Three main models were employed in this study: Random Forest, Long Short-Term Memory (LSTM), and a model enhanced with SMOTE. The evaluation was conducted using accuracy, precision, recall, and F1-score metrics, supported by confusion matrix analysis. The results indicate that the model integrated with SMOTE outperforms the others, achieving an accuracy of 0.5447, precision of 0.5512, recall of 0.5447, and F1-score of 0.5462. This model also demonstrates a more balanced classification performance compared to the LSTM and Random Forest models. These findings confirm that the application of oversampling techniques such as SMOTE, when combined with appropriate algorithms, can significantly enhance predictive performance in imbalanced datasets. The study contributes to the development of predictive models for commodity prices based on historical data and opens opportunities for further exploration of more adaptive hybrid methods in future research.]]>
Copyrights © 2025
