Research On Sales Forecast Of Entity Retail Industry Based On Data Mining

Accurate sales forecast can help enterprise managers effectively formulate business plans,thus making business strategies and work plans more effective and accurate,and improving the comprehensive competitiveness of enterprises.At present,time series analysis and machine learning modeling are used to predict retail sales.The prediction method of time series analysis highlights the role of time factor in the prediction,and does not take the external factors into account,so when the external environment changes greatly,the predicted value will often have a large error.The machine learning modeling prediction method analyzes all the factors affecting the prediction,and the prediction method is flexible and efficient.However,the independent machine learning modeling prediction method is based on the presupposition,which has limitations,leading to the fact that a single model cannot contain all the information in the data,and it is necessary to combine multiple models to build a mixed model for prediction.The research focus of this paper is to build a mixed model based on the real sales data to improve the prediction accuracy and generalization ability of the prediction model.The research object of this paper is the sales data and store information d ata of multiple stores of German brick-and mortar retail giant Rossmann company.The exploratory data analysis is carried out on these data,and the impact of various variable s on store sales is analyzed and visualized.According to the characteristics of the data set,the data preprocessing is carried out and the characteristic engineering is built.The performance of the linear regression,random forest,support vector machin e and XGBoost model in the sales forecast is compared.In order to further improv e the accuracy and generalization ability of XGBoost model in sales forecasting,a hybrid model pca-k-means-xgboost based on clustering algorithm and XGBoost model is designed.The mixed model includes principal component analysis,k-means clustering and XGBoost regression,selects the optimal clustering class number after dimensionality reduction of the feature set,trains the XGBoost model with different class data after clustering,and makes the final prediction after the mixed model.Experimental result s show that the PCA-K-Means-XGBoost hybrid prediction model further improves the prediction accuracy and generalization ability.Compared with the method of building a single model for prediction,the mixed model can contain all the dimensional information of the original data set and is more suitable for the prediction field with a lot of information,such as retail sales forecast.