Modelos LSTM para predição do preço da soja com base em dados climáticos brasileiros

Abstract

The importance of Brazil in the agricultural scenario is unquestionable, being one of the largest soy producers in the world. With a growing productivity, Brazil has achieved more and more prominence in the world scope. The climate actively influences soy production, however, it is worth remembering that Brazil is a continental country. For this reason, climate data differ from one region to another. The objective of this work is to carry out a predictive analysis of soybean prices using Brazilian climate data and to identify the influence that Brazil’s climate has on the international price of soybeans. For this, a climate database and complementary data were obtained, such as the value of the dollar, inflation and the annual production of soy. These data were collected from 2000 to 2020 from the entire Brazilian territory. For the experimental evaluation, climate data from the states with the highest soy production were chosen, namely Mato Grosso, Mato Grosso do Sul, Goiás, Paraná and Rio Grande do Sul. After the pre-processing of these data, predictive models were built using the Long Short Term Memory (LSTM) method. So, the method was applied to each state and two different configurations (one with climate data and the other without climate data), in total ten models were built. In the performance evaluation, the Root Mean Squared Error (RMSE) and the Mean Absolute Error (MAE) were calculated for each model. In this case, the models built using Paraná data achieved the best results for both configurations in the two metrics. After that, statistical hypothesis tests were applied and none of them showed a statistically significant difference between the models, that is, the use of Brazilian climate data does not significantly influence the international price of soy. As much as the results have shown that Brazilian climate data had no influence on the price of soybeans in the experimental evaluation, this work had several contributions, such as the creation of a climate database for use in other studies and the predictive analysis of the price of soybeans, enabling help for producers in decision-making.