查看文章

Sales forecasting is an important task for managers to make replenishment according to historical sales. A flexible and easy to use forecasting solution will benefit retailers from loss of sale, over supply and merchandise waste. Deep learning is a popular topic in many fields in recent years. This paper tests a long short-term memory (LSTM) recurrent neural networks (RNN) on 45 weeks point of sale (POS) data of 66 products without considering the impact of seasonality and promotions. One fourth of products have a relatively low forecasting error, which validates the feasibility of the LSTM network to some degree.