Performance Evaluation of Ensemble Learning Stacking in Rice Price Prediction

Rice prices in Indonesia play an important role in maintaining economic and social stability. However, unpredictable price fluctuations often compromise people purchasing power and national food security. This challenge is compounded by uncertainties in production, population growth, and climate change such as EI Nino that affect rice production and distribution. In addition, the limitations of traditional prediction methods often result in inaccurate price forecasts, making it difficult for policymakers and market players to make timely decisions. In this context, machine learning has great potential to improve the accuracy of rice price prediction. To address this issue, this research applies the Ensemble Learning (EL) Stacking method using Random Forest (RF) as the meta learner and Recurrent Neural Network (RNN), Least Square (LS), and Long Short-Term Memory (LSTM) models as a basis learner. Evaluate indicates are used to measure such performance. As a result, EL Stacking significantly improved prediction accuracy, with Coefficient of Determination ($\mathbf{R}^{2}$) 0.9996 and very low error Mean Absolute Error (MAE) 9, Mean Squared Error (MSE) 288, Root Mean Squared Error (RMSE) 16, and Mean Absolute Percentage Error (MAPE) 0.0007, on a 60:40 percentage ratio dataset split. This study demonstrates the potential of EL Stacking in predicting rice prices with high accuracy, helping stakeholders maintain price stability and strengthen national food security.