Research On E-commerce Sales Forecasting Method Based On Deep Learning

In the past 10 years,the domestic e-commerce market has prospered in the rapid development of the Internet industry,and e-commerce companies have experienced a"golden age" of rapid growth.However,in 2016,the growth rate of the domestic online retail market showed less than 30%for the first time.The low natural population growth rate gradually led to the disappearance of the traffic dividend.This market is also shifting from "blue ocean" to "red sea".Refinement and intelligent operation have become the key to the growth of e-commerce companies.Predictive work is an important part of the e-commerce intelligent operation.Accurate forecasting will have a profound impact on the company's production,marketing and logistics decisions.From a macro perspective,enterprises can reduce the"bullwhip effect" of supply chain information transmission,accurately grasp the market size and future demand trends of products;from a micro perspective,accurate forecast results can be applied to purchasing decisions,commodity pricing,and warehouse front-end,targeted marketing,personalized customization and other scenarios to help companies better win customers and markets.Data mining technology is widely used in e-commerce enterprises,but the reality of business and prediction technology is still a realistic challenge for e-commerce companies.First of all,this paper starts from the business reality of e-commerce enterprises,based on the theory of consumer purchase decision-making process to refine various factors affecting e-commerce sales,and further summarizes the characteristics construction of e-commerce sales forecast,which provides reference for practical application of enterprises.Secondly,this paper proposes a deep learning-based e-commerce sales forecasting model,which combines feature learning of convolutional neural networks with Xgboost regression model fusion prediction.Finally,based on the data of E-commerce company A,this paper compares the effects of linear model,random forest and other tree models with CNN_XG fusion model.It is found that,firstly,the nonlinear model is much higher than the linear model in terms of prediction accuracy.Secondly,the artificial feature work has certain help to improve the prediction accuracy of the model,but the effect is limited.Third,the basis of this paper is based on The deep learning CNN_XG fusion prediction model can have better prediction performance on a small number of feature sets,indicating that feature learning can efectively reduce the feature engineering work in the actual business.The convolutional neural network can effectively utilize the time series data in the basic features.The characteristics give Xgboost more information,and ultimately improve the accuracy of the overall prediction.The e-commerce sales forecast feature engineering construction method and the deep learning-based fusion prediction model proposed in this paper can improve the accuracy of the artificial feature engineering and improve the accuracy.It also proves that this paper has some reference value in the practice of the enterprise.