| In modern society,the new retail industry is booming and the traditional retail industry is facing great challenges.In this context,it is particularly important for the traditional retail industry to use machine learning methods to make reasonable predictions about future merchandise sales.This article conducts a comprehensive and thorough analysis of the1941-day U.S.Wal-Mart supermarket merchandise sales data from January 29,2011 to June19,2016,effectively improving the scientific nature of the merchandise sales forcast of large-scale supermarket chains,and provides practical solutions for commodity sales forecast of retail companies.First,perform data source analysis,data exploration and visualization,and data preprocessing on Walmart supermarket merchandise sales data.It is found that the data is hierarchical time series and the importance of each feature is mastered.At the same time,the data is converted into the data form used for supervised learning.Secondly,in terms of feature construction in feature engineering,in addition to using common methods to extract basic features,it also uses feature construction methods specific to time series,such as introducing lagging features,trending features,and introducing features through extended windows.Then use Random Forest,XGBoost,and Light GBM respectively to model the training samples according to the supermarket dimension,and make predictions on the daily sales of all commodities in the next 28 days.In the context of tens of millions of data and multi-features,in terms of model traning time,Light GBM has significant advantages compared to Random Forest and XGBoost.The training time is 1/15 and 1/5 of the former two,respectively.In terms of model prediction accuracy,Light GBM also has advantages compared to Random Forest and XGBoost,and the prediction errors are reduced by 15.094% and 7.898%,respectively.To sum up,the Light GBM model can be better used to deal with the problem of largescale supermarket merchandise sales forecasting under the background of massive data and multi-features,and has great significance in improving the economic benefits of large chain retail enterprises. |
