COMPARATIVE PERFORMANCE OF SARIMAX AND LSTM MODEL IN PREDICTING IMPORT QUANTITIES OF MILK, BUTTER, AND EGGS

Abstract

This study evaluates how well SARIMAX and LSTM models predict monthly imports of HS-04 commodities (butter, eggs, and milk) in Indonesia. Data were provided by BPS Statistics Indonesia, Bank Indonesia, Ministry of Trade, Trade Map, and Indonesia National Single Window and used from January 2006 to February 2024. The SARIMAX model included exogenous variables such as inflation rates, USD/IDR exchange rates, and major Indonesian holidays (Eid al-Fitr, Eid al-Adha, Christmas, and Lunar New Year). The results show that the SARIMAX and LSTM models predict the import volumes of butter, eggs, and milk with good accuracy. However, the SARIMAX model demonstrated superior forecasting accuracy, achieving a lower RMSE of 7547.89 and a MAPE of 13.16 compared to the LSTM model, which yielded an RMSE of 8787.73 and a MAPE of 14.89. The SARIMAX model performed significantly better when the lunar new year was added as an exogenous variable. In order to support price stability and economic growth in Indonesia, this research provides policymakers and industry stakeholders with critical information to optimize import management strategies for butter, eggs, and milk commodities. Accurate forecasts can contribute to price stability, enhanced food security, and sustainable economic development in Indonesia by enabling informed decisions on import quotas, tariff adjustments, investment in domestic production, and strategic reserves.