LSTM forecast of volatile national strategic food commodities

Abstract

Using the Long Short-Term Memory (LSTM) forecast, this study suggested a short-term projection model for national critical food pricing commodities. The model was trained using historical time-series data from each commodity price over the previous three years. The results demonstrated that the proposed LSTM architecture model was generalizable to all commodities and performed well in the majority of cases. This result indicates that the model is resilient and can be used to forecast commodity prices and offer accurate forecasts for most of the ten volatile national strategic foods, with an error value of less than 0.01 and an accuracy value of >95%. The model, however, failed to recognize the pricing pattern in cooking oil and beef commodities, both of which had increasing trend patterns. This shows that the model may be unable to effectively estimate commodity prices in the face of fast price fluctuations. The magnitude and quality of the dataset hampered the investigation. The time period selected also influenced the study. Future research should employ a more extensive and diversified dataset to increase the model's performance, allow it to learn more patterns and make more accurate predictions, and could use a more extended lookup date to improve forecast accuracy. This would enable the model to account for more recent pricing changes. Despite the limitations, the results of this study are promising and could be used to develop a more accurate and reliable food price prediction model. Policymakers and stakeholders could use the model to make informed food prices and inflation decisions.