The following paper has the objective to develop forecast models, critical aspect in companies. The optimization of the forecast brings, in fact, advantages in the organization and management of the entire supply chain. The use of machine learning can provide a valid substitute or support for traditional forecasting models and a backing for business decisions, bringing process innovations to companies. Machine learning models have been developed on R, statistical software, and subsequently integrated on already existing business planning systems. Two methodologies have been studied and compared: ARIMA, a linear and more traditional model based on regression and Neural Networks, which allow the implementation of non-linear models by gathering relationships between data and predicting by analogy.