Uso de algoritmos para predição de séries temporais: um estudo das plataformas Facebook's Prophet e Amazon DeepAR

Abstract

Nowadays, with advances in Artificial Intelligence, and the increase in processing capacity, machine learning, neural networks and deep learning have become able to make predictions of the most varied time series. This work presents concepts of Artificial Intelligence, Machine Learning and especially the time series prediction platforms widely used at the moment: Facebook’s Prophet and Amazon DeepAR. In this paper, software of both aforementioned platforms are created in order to predict the value of gold (oz/USD) through them. It presents a state-of-the-art reference and the main technologies used, works related to the theme, the paper carried out on each platform as well as it’s results: using the MAPE error metric, the Amazon DeepAR software obtained a rate error of 0.19% and Facebook’s Prophet of 0.006%, this one proving to be more suitable for this type of prediction. Finally, considerations relating to the results found are presented with ideas for future work to be carried out.