Study of E-commerce Sale Prediction Based on Machine Learning Methods

Rong  Liu; Annie Anak  Joseph; Wanzhen  Wang; Feng  Cao; Na  Li; Taihao  Zhang

Authors

Rong Liu Department of Electronic Information, Qilu Institute of Technology, Jinan, PR. China
Annie Anak Joseph Department of Electrical and Electronic Engineering, Faculty of Engineering, Universiti Malaysia Sarawak (UNIMAS), Kota Samarahan, Sarawak, Malaysia
Wanzhen Wang Department of Electronic Information, Qilu Institute of Technology, Jinan, PR. China
Feng Cao Department of Electronic Information, Qilu Institute of Technology, Jinan, PR. China
Na Li Department of Electronic Information, Qilu Institute of Technology, Jinan, PR. China
Taihao Zhang Faculty of Computer Science and Technology, Qilu University of Technology, Jinan, PR. China

Keywords:

Sales forecasting, time series data, model performance comparison, practical, metrics

Abstract

Precise overall sales forecasting is essential in the sales domain for controlling slow-moving commodities and cutting inventory expenses. However, seasonality, trends, and multi-product scenarios provide challenges for established methods of sales forecasting. For time series data and complicated patterns, models such as gated cycle unit (GRU), recurrent neural network (RNN), and short- and long-term memory network (LSTM) were chosen to increase processing power. To find the best models for sales forecasting, the performance of these models was compared using metrics (MAE, RMSE, and ). It is found that GRU model is the best model in this field. In order to assure the research's suitability from a scientific and practical standpoint, these additional components have been added to increase the study's scope, address the issue of previous research using these models sparingly or not at all, and look for more efficient ways to forecast sales.