MODELING AND PREDICTING INDONESIA RICE PRICES USING HYPERPARAMETER OPTIMIZATION XGBOOST

Abstract

This study explores the application of the XGBoost model, fine-tuned through hyperparameter optimization and cross-validation, for forecasting rice prices in Indonesia for the two years following July 2024. Utilizing a comprehensive dataset spanning from January 2010 to July 2024, the research emphasizes the importance of detailed feature engineering, including temporal and cyclical patterns, to enhance the model’s predictive accuracy. A 4-fold cross-validation approach was employed, resulting in a rigorous evaluation process that involved over 26,000 model fits. The study highlights the effectiveness of XGBoost in capturing complex patterns within the data, yielding highly accurate predictions. Additionally, the integration of external factors—such as climate conditions, government policies, global market trends, economic indicators, and technological advancements—is recommended to further refine the model and ensure adaptability to real-world conditions. The findings suggest that this approach provides valuable insights for stakeholders, including policymakers and market analysts, facilitating informed decision-making regarding production, pricing strategies, and food security. This research underscores the potential of advanced machine learning techniques in improving time series forecasting within dynamic and complex markets like rice pricing.