Previsão de demanda aplicada a séries temporais da vitivinicultura gaúcha

Abstract

The agricultural industry is characterized by a complex environment and predictive models can help managers and farmers in organizing and planning their businesses. In Brazil, vitiviniculture has gained strategic space in recent years. The aim of this study was to forecast demand for the production and sale of wines in state of Rio Grande do Sul. For this, we used the Cross Industry Standard Process for Data Mining (CRISP-DM) and the Amazon Forecast platform that applied conventional and advanced algorithms for forecasting for a period of the time series of vitiviniculture in Rio Grande do Sul. The ARIMA method proved to be the best predictive algorithm for meeting the defined criteria for precision. Forecast demand for wine production was similar to actual demand, while forecast demand for wine sale was lower than actual demand, which can be attributed to unexpected effects in the last year such as favorable weather events and COVID-19 pandemic. The approach used in this study is comprehensive and compatible with different contexts, and can be used to forecast demand for any product that has records in a time horizon.