Design And Implementation Of Medical E-commerce Platform Based On Data Mining

In recent years,China’s pharmaceutical e-commerce industry has developed rapidly,and due to the current impact of the new crown epidemic and the introduction of favorable government policies,the development of pharmaceutical e-commerce has once again moved to a new level.At present,the pharmaceutical e-commerce industry has entered a critical period of development,but there are still many problems in the cognition of the function of the platform and the positioning of the sales target of the pharmaceutical platform by the developers of the pharmaceutical e-commerce platform and the merchants who enter the pharmaceutical platform.Many pharmaceutical companies face the risk of continuous loss and bankruptcy in the increasingly fierce competition due to their vague positioning and unclear strategic layout.Predicting the sales amount of merchants can help merchants plan in advance to avoid the occurrence of such problems,but the traditional forecasting algorithm is relatively single,the generalization ability is weak,the prediction effect is poor,and it is impossible to make accurate predictions about the sales of merchants.Therefore,on the basis of completing the development of the pharmaceutical e-commerce platform,this paper proposes to use stacking fusion strategy to predict the sales situation of merchants.The pharmaceutical e-commerce platform mainly includes the construction of the user side and the merchant side.The user side mainly has the home page,shopping cart and my and other modules;the merchant side mainly has the business home page,product management,transaction management,revenue center,message center and other modules.Merchant sales data is provided by a pharmaceutical e-commerce company M,this paper uses the Stacking strategy to use the time series model ARIMA,LSTM,XGBoost as the primary learner,and the LightGBM model as a secondary learner to predict the number of new users,sales amount and order volume of the merchant.Experimental results show that the prediction of the number of new users,sales amount and order quantity using stacking fusion strategy is 1.08%,0.74%,0.82%,0.81%,0.32%,0.55%,1.05%,0.63% and0.81% respectively compared with the average absolute percentage error of ARIMA,LSTM and XGBoost,which are reduced by 1.08%,0.74%,0.82%,0.55%,0.81%,respectively.Finally,after experimental verification,the Stacking strategy proposed in this paper can concentrate the advantages of each algorithm model,improve the accuracy of sales prediction of merchants,and is of great significance to merchant decision-making.