Predicting the tide: A deep-learning approach for understanding the whitemouth croaker prices in Northeast Brazil

Abstract

Seafood represents the most traded animal protein globally, with a significant contribution to food security in emerging economies. Therefore, it is crucial to conduct studies that aim to predict fluctuations in price to ensure the affordability of these products. Such studies could inform the establishment of political measures designed to minimize large variations in prices. In this context, the present research aimed to evaluate the historical price series, trends and seasonality of the whitemouth croaker traded in a supply center in northeastern Brazil. In addition, we tested the predictability capacity of a long short-term memory (LSTM) neural network in the context of seafood economic analysis. The prices exhibited a general upward trend, with occasional declines, and a more pronounced seasonal impact in recent years. The LSTM demonstrated low error scores of root mean square error, mean absolute error and mean absolute percentage error, indicating its suitability as a tool for monitoring the fluctuations in commodity prices. Nevertheless, adhering to certain standards is essential to prevent erroneous predictions that could result in misguided policy decisions.