<![CDATA[This study develops a multi-horizon Long Short-Term Memory (LSTM) model to forecast food commodity prices in Indonesia, supporting the national Free Nutritious Meal (MBG) program. The MBG initiative increases daily and annual food demand for over 50 million beneficiaries, affecting rice, meat, eggs, vegetables, and cooking oil. Historical price data were used to train the model, which effectively captures nonlinear temporal patterns and seasonal variations. Forecast results indicate heterogeneous volatility patterns across commodities. Staple commodities such as rice and cooking oil exhibit relatively stable trends with minor monthly fluctuations, whereas perishable and climate-sensitive commodities such as chili, eggs, and sweet potatoes demonstrate higher short-term volatility. To ensure temporal robustness, the model was evaluated using a Rolling Forecast Origin (RFO) validation strategy with an expanding window approach. Evaluation on the test set yielded a MAE of 2,108.37, RMSE of 3,933.30, and MAPE of 6.61%, indicating approximately 93% accuracy. The model is implemented in an interactive interface allowing users to select commodities, set prediction horizons, and export results, providing a practical tool for procurement planning, distribution management, and national food security.]]>
Copyrights © 2026
