Research On The Sales Forecasting Of Proprietary Commodities Of Cross-border E-commerce

Cross-border e-commerce,as a new form of foreign trade,has great potential for development.Although the prospects for cross-border e-commerce development are good,the management of global supply chains is of crucial importance in order to obtain a a space for one person and steady development in the fierce competition.Compared with domestic e-commerce,cross-border e-commerce is more complicated in process,farther in space,and longer in logistics time,which puts forward higher requirements on its supply chain management.Purchasing is the source of supply chain management and sales forecasting is the basis for developing procurement plans,so an accurate sale forecast can better guide procurement and improve supply chain efficiency.The research question of this paper is how to improve the sales forecast accuracy of X company's own products.Through interviews and reading the literature,it can be found that X company has two main problems of forecasting commodity sales,such as single factor and subjective qualitative prediction.The low accuracy of sales forecasting results in a backlog of some of the company's commodity inventory and the fact that another part of the products is often out of stock,which not only increases the cost but also reduces the company's market competitiveness imperceptibly.Because the purpose of sales forecast is to develop a procurement plan,this paper takes sku as the forecast unit's monthly short-term sales forecast.In this way,the forecast is more practical.The main research results of this paper are as follows:Firstly,according to the theory of online consumer behavior,relevant literature review and actual data of the company,10 impact factors are initially selected from three dimensions: users' behavior,commodity information and transaction behavior.In addition,this paper makes use of the correlation analysis to eliminate the low correlation.Finally,this paper selects 7 impact factors such as clicks,additional purchases times,collection times,discounts,brands,categories,sku historical sales as the characteristic variables when the factors function as the model.Secondly,this paper constructs the GBDT sales forecasting model based on the above seven influencing factors and the ARIMA sales forecasting model only using historical sales.The model prediction results are compared and evaluated through the actual test set.The results show that the average relative error of the GBDT model is 8.93% lower than the 23.77% of the ARIMA model,and both are much lower than the current average relative error of X company 41.88%.Through further analysis of the weights of the characteristic variables of the trained GBDT model,it can be found that the three variables which have a greater impact on the prediction results are the historical sales volume,the number of purchases,and the number of collections.This shows that this paper considers multiple impact factors and selects appropriate prediction methods so as to effectively improve the accuracy of the company's product sales forecast and to better formulate procurement plans and optimize inventory structure,which has certain meaning for the actual operation of the company.