Statistical and Machine Learning-based E-commerce Sales Forecasting

Wenxiang Dong, Qingming Li, H. Vicky Zhao

Abstract: Market share analysis and sales forecasting have always been an important research area. It is of great significance to predict sales through existing information and provide guidance to merchants and markets to obtain higher profits. However, most of the traditional research focuses on brick-and-mortar retail stores, while few works studied E-commerce markets. In this paper, we use the historical data in the e-commerce market to establish the model to predict the sales. According to the characteristics of different data, three types of prediction models are: Incentive Auto Regressive Integrated Moving Average(I-ARIMA), Long Short-Term Memory(LSTM) and Artificial Neural Network(ANN). These three methods can deal with the problem with different accuracy requirements and different data types. This paper studies the advantages and disadvantages of the three types of models on different data sets, and provide important guidelines to merchants on their marketing strategies.