Application of Machine Learning Model and Hybrid Model in Retail Sales Forecast

Abstract

Using time series data to predict future sales changes of products is of great significance to every retailing company in terms of management and planning of resources. In order to find an effective method to improve the accuracy of sales forecasting of retail goods which strongly influenced by season and holiday, this paper analyzes the feasibility of traditional time series model, hybrid models based on time series model and machine learning model, and machine learning model in predicting Walmart sales. The Prophet model which decomposes trend, season, and holiday and the machine learning model-lightGBM model- are used to train and test Walmart supermarket sales data from 2011-01-29 to 2016-06-19, and use data from 2016-06-19 to 2016-08-14 to make prediction and empirical analysis. The results suggest that the Root Mean Square Error (RMSE) of the Prophet model and the LightLGB model are 0.694 and 0.617, respectively, indicating that the machine learning model performs well in the sales forecast of retail stores. This provides a new idea for retailers to forecast sales by category and region.