<![CDATA[Introduction/Main Objectives: Rice is a key staple commodity influencing food security and inflation in Indonesia, making accurate price forecasting essential. In this study, we aim to compare ensemble boosting models and identify the best-performing model for rice price prediction. Background Problems: Notably, rice prices exhibit non-linear patterns over time, while classical statistical methods have limitations in capturing such complexities, resulting in suboptimal forecasting performance. Novelty: This study proposes a lag-based approach that uses lag variables as the only predictors, arranged across multiple input schemes to flexibly capture historical patterns without external variables. Research Methods: Daily national medium rice price data (Jan 2021–Jan 2026) from the National Food Agency are modeled using Gradient Boosting Machine (GBM) and LightGBM, with hyperparameter tuning via Optuna. The forecasting framework relies exclusively on significant lag variables without incorporating exogenous factors. Model performance is evaluated using RMSE, MAE, and MAPE. Findings/Results: LightGBM with optimized hyperparameters achieves the best performance (RMSE = 66.389; MAE = 50.213; MAPE = 0.362%). Furthermore, forecasts for the next 89 days indicate stable prices around Rp13,360–Rp13,395/kg, with no significant fluctuations.]]>
Copyrights © 2026
