APPLICATION OF RECURRENT NEURAL NETWORKS IN FRUIT PRICE PREDICTION IN THE AGRICULTURAL SECTOR OF BAHIA

Abstract

This study investigates Recurrent Neural Networks (RNN), specifically LSTM and GRU, in the price forecasting of fruits grown by family farmers in Bahia, based on 1,883 records of banana nana, banana prata, and papaya formosa. The LSTM model presented the best performance, with RMSE results ranging from 0.186 to 0.606, MAE from 0.142 to 0.483, MAPE from 7.286 to 16.624, and MSE from 0.035 to 0.367 for the fruits analyzed. The potential of RNNs in supporting decision-making in the agricultural sector is highlighted, with proposals for future work that include incorporating exogenous variables and the development of a free platform for small producers.