Study On Sales Forecasts Of Stores Based On Xgboost Method

Reliable sales forecasting enable help store managers to create effective staff schedules and stay focus on what is the key factors for clients and the team. So as to improve the production mode and increase productivity and motivation. By create a robust prediction model will help store managers stay focused on what's most important to them.This paper is take the sales and store information data set of 1115 stores provided by famous German company Rossmann as the objective of data mining, using the Exploratory Data Analysis and visualization technology,implement with Python and R programming language. To compare with the performances of Xgboost(Extreme Gradient Boosting), Random Forest, Glmnet(Lasso and Elastic-Net Regularized Generalized Linear Models) and LM(Linear Model), TSLM(Time Series Linear Model) in sales forecasting, then preliminary finding showed Xgboost model achieve good results in RMSPE(Root Mean Square Percentage Error) evaluation criteria.To further improve Xgboost method in accuracy and generalization performance of the sales forecasting, the paper combined with Feature Engineering, and using Emsemble Learning method, to optimize the Xgboost model by tuning parameters. The experimental results show that the final models of using Glmnet and Xgboost model fit residuals and combining the advantages of LM and TSLM in trend and seasonality can improve the accuracy and generalization.The ensemble models based on Xgboost applies not only to predict the German retail sales, but also extend to the domestic retail industry and electronic business platform. It is significant meaning to improve the operation mode, commodity price, distribution and corresponding precision marketing of stores.