Sales Forecaster: A Bis-Attention LSTM Encoder-Decoder for Local Store Sales Forecasting
Download as PDF
DOI: 10.25236/icemeet.2021.032
Author(s)
Ziye Zhou, Zhiying Chen, XuYUAN
Corresponding Author
Ziye Zhou
Abstract
Time series forecasting is deeply and broadly studied throughout the years with various applications in multiple industries, such as finance, weather, and environment. Deep neural networks combined with attention mechanism is capable of capturing long-term patterns while effectively incorporating information from other variables. This paper applies and modifies a few of the deep neural networks based on attention mechanism for a real-world problem, local store sales forecasting, in a theoretical manner. Apart from the effect of long-term dependencies, this model is able to take dynamic spatial correlation into account, which is widely encountered in local store sales forecasting.
Keywords
Time Series, Forecasting, Attention, Neural Network, Sales