Sales prediction using Transformer-based models

In the last years, deep learning algorithms that have demonstrated enormous potential in NLP, have been studied in various field such as time series forecasting. In particular, state-of-theart approaches such as Seq2Seq and Transformer models have been proven to be suitable for temporal data. In this internship project, the main goal is to implement and investigate the application of these models on the sales prediction task. The proposed models are trained and evaluated on a retail dataset given by the internship company. The results obtained demonstrate that these model are able to handle the task but, in some cases, the prediction performance does not overcome the simpler baseline models.