Revisão bibliométrica sobre modelagem e previsão do preço da soja: uma comparação entre os modelos ARIMAX, redes neurais e máquina de aprendizado extremo

• Ano de defesa: 2021
• Autor(a) principal: BRUNO MATOS PORTO
• Orientador(a): Matheus Wemerson Gomes Pereira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Fundação Universidade Federal de Mato Grosso do Sul
• País: Brasil
• Palavras-chave em Português: Desempenho de previsão, Previsão de séries temporais, Commodity agrícola
• Link de acesso: https://repositorio.ufms.br/handle/123456789/3677

Forecasting models are useful in the prediction of prices commodity agriculture and serve for decision-making from rural producers, investors, agricultural authorities, and market participants. The scientific gap on price forecasting commodity agriculture was identified through bibliometric analysis, which generated results in quantitative parameters of scientific production and showed trends on the subject, in addition, provided references for econometrists in the field of agricultural price forecasting commodities agriculture. The forecasting problem was to model a set of monthly soybean price data in three states: Paraná, Rio Grande do Sul, and Mato Grosso. The general objective was to evaluate the models autoregressive integrated moving average with exogenous variable (ARIMAX), multilayer perceptron neural network (MLP), and extreme learning machine (ELM). The theoretical model of the research was the law of the single price. The training and test sets were modeled using methods ARIMAX, MLP, and ELM. Subsequently, the forecasting in the test sets is compared with forecast performance indicators. The forecast accuracy of the methods was compared through evaluations of the box plots and tests of variances of the errors of each model, thus identifying whether there were significant statistical differences between the error metrics generated in the forecast. The results reveal that ARIMAX is the best approach to forecast prices in Paraná, the MLP obtained the performance of the superior forecast, to predict prices from Rio Grande do Sul. Finally, MLP and ELM with the best accuracy among the comparisons from Mato Grosso's price forecasting. The other comparisons did not have significant differences (p> 0.05) between the variances of the errors of the forecast models. From this analysis, the study concluded that, for the prediction of soybean prices, artificial intelligence (AI) models, in some cases, are not superior in comparison to the statistical model ARIMAX.