Walmart Sales Prediction Based on Machine Learning

Abstract

Accurate sales forecasting can improve a company's profitability while minimizing expenditures. The use of machine learning algorithms to predict product sales has become a hot topic for researchers and companies over the past few years. This report features the machine learning sales prediction model that combines the ML algorithm and meticulous feature engineering processing to predict Walmart sales. The following regressions are analyzed in this paper: linear regression, random forest regression, and XGBoost regression. The regression analysis has been tested for the same time period every year for three years from 2010 to 2012 on a continuous time basis. The experiments show that XGBoost algorithm overperforms the other machine learning methods by examining the same evaluation metric (WAME). The findings can contribute to a better understanding of the development of new decision support for the retail industry e.g., Walmart retail stores. Moreover, this paper also represents a detailed procedure to rank the feature importance for the dataset. Within the next few years, the ML algorithm is destined to become an important approach for business forecasting. However, this strategy largely ignores the time series method for accuracy.