Study On The Behavior Of EC Customers Based On Machine Learning

With the rapid development of Internet technology,network is closely related to people's life,from the growing number of Internet users to the developed rapidly express industry,all show the network brings people convenience.Nowadays almost everyone in the case of online shopping,it is not only an opportunity for electric business platform,but also a big challenge.In the increasingly competition of the electricity business environment,who can better to retain customers,to attract new customers is the key of electric business platform development.under the background of network technology and artificial intelligence,at the end of the day,electric business platform competition is the technology competition,who can use the advanced technology platform for yourself more intelligent,more convenient,recommend more accurate,etc.,who will be able to have a place in the field of electricity.For EC research in the final analysis is the study of user behavior data,in the vast amounts of user behavior data,how to dig out the potential value of the user information,explore the user's needs and behavior rule,etc.,is the key of the user behavior research.Nowadays,the user behavior research has become a study hotspot at home and abroad,and machine learning in artificial intelligence applications in the field of e-commerce.This paper studies the relation between EC customers and customers' orders,and using the machine learning algorithm in artificial intelligence to learned customer behavior modeling,and combining with the ecological classical ant colony algorithm of swarm intelligence algorithm optimize the xgboost algorithm of machine learning algorithms.in the model framework we use the study method,and compared xgboost algorithm with the other machine learning algorithms include logistic regression(LR),support vector machine(SVM),random forest algorithm based on decision tree algorithm(RF).it is shows that the optimized xgboost algorithm is better than others in EC customers study,and the prediction accuracy is above 85%.Before building model framework,this paper first to get the customer behavior data of B2 C e-commerce sites in exploratory data analysis,users in the early morning of interaction is higher,and most of these users are aged 26 to 35 young people,in the users order behavior purchase rate to decline over time,etc.;To characteristics of engineering construction,according to the results of the analysis of preprocessing,feature selecting fits the characteristics of research;And characteristics of the processing according to the user's behavior over time to join the IDF values,making the edge features much easier to get.Based on time dimension to the training set,according to the change trend of time,time sliding window is used to the characteristics of training and testing,training and test for 5 days period,through continuous increase in the number of moving window makes the training set data,and according to the result of each slide,adjust the parameter weights,so that the prediction accuracy.Experiments show that the optimization XGBoost algorithm has the best prediction in the study.