Retail Time Series Prediction Based on EMD and Deep Learning

摘要

The paper presents a new method based on EMD and deep learning. The method transforms the retail time series prediction problem into a signal analysis processing problem by using “divide and conquer” concept. It can decompose the source time series and build the suitable deep learning models for each sub-sequence by using the suitable retail time series feature. The experiment shows that the effect of the method performs better than those traditional machine leaning methods and single deep learning method on MAE and RMSE indicators in the retail time series prediction task. The method can be using to conduct the retail business behavior for a better sales result.