Research Of Sales Forecast Of E-Commerce Company Based On Data Mining

Since 2016,revenue growth of JD.com and Alibaba.com has dropped from 40%to 50%.Compared with the past growth rate of the three digits,this indicates that the growth rate of e-commerce revenue has tended to be flat,and bonuses brought by the traffic and mobile internet have been over.The ultra high-speed growth has basically come to an end.The customer needs are more varied,the competition among enterprises is more fierce,and the e-commerce enters the refined operation stage.A consensus has been reached in the enterprise and society,application of new technologies cam improve the company's operating standards,especially data science and technology.Data Mining technology has been applied to the world's leading enterprises since the beginning of the 21st century.Domestic enterprises began to attach importance to it around 2012,but at the time they were constrained by the lack of large amounts of data.After many years of data platform construction and accumulation,many companies now have relatively rich historical data,and it can be said that the accumulation of theory,tools,and raw materials has basically matured.Companies need to make effective predictions on many data in their business areas and operations to arrange for the next phase of production,logistics,marketing,and expansion of new businesses.Through the study of literature and business surveys,it is found that the biggest problem that predictive technology has in enterprise applications is that the forecasting needs of the business are out of touch with the forecasting technology.This article uses the Cross Industry Process for Data Mining(CRISP-DM)methodology to guide forecasting tasks and introduces data-centric knowledge discovery tools into traditional demand forecasting.Not only has the technology of data mining been introduced,but also the more important and useful value is the introduction of the data mining process.From the perspective of consumer shopping behavior and e-commerce network marketing strategy,a comprehensive summary of the models and influencing factors affecting sales forecasts.It provides comprehensive business knowledge for the establishment of data mining models,and also guides the acquisition and analysis of relevant data in the data mining process.This is the missing or weak link in previous studies.This paper adopts the method of causality prediction to establish three most common data mining models,such as random forest,support vector machine and neural network,to make predictions.In addition,it compares classical multiple linear regression algorithm.The ten-fold cross validation method was used to evaluate the performers of these models.It was found that under the current data set,the support vector machine with radial root had the best prediction effect,reaching an error rate of 7%,and the generalization performance of the model was most stable.However,compared with S VM,the performance of multiple linear regression is also good,and the prediction accuracy is close.It shows that under the current business scenario,the influence of the influencing factors on sales volume is nearly linear.Under the guidance of the current methodology,excellent prediction results and rich management implications can be obtained,which proves that the proposed forecasting process has greater application value.Data inputting,processing,analysis,and modeling in the study are all implemented by R language program,an open source software.