Análise comparativa entre técnicas de aprendizado de máquina aplicadas para a predição de preços de produtos hortifrutícolas

Abstract

Family farming is characterized as any form of land cultivation managed by a family, employing its own members as the main labor force. Most agricultural establishments in Brazil fit this definition, however, the area occupied by these farmers and the available resources are much smaller compared with that of large agricultural companies. With limited area and resources, it is extremely important that small farmers are able to optimize their production to ensure the sustainability of their business. The objective of this work is to study and compare the performance of different machine learning techniques in the task of predicting time series of horticultural products prices for short term periods, in order to help small farmers to negotiate a fair price for their merchandise. From the results obtained by the induced models, it was possible to conclude that the Random Forests, Support Vector Regression and Long Short-Term Memory techniques have a similar performance in the task of predicting prices of fruits and vegetables. In addition, the models have the potential to be used as tools capable of assisting producers when selling their merchandise.