Ensemble Learning And The Application Of Predicting User Churn On E-commerce

Nowadays,the e-commerce industry is developing rapidly,and users have been essential for survival and development of business enterprises.The survey found that there existed widespread loss of users on e-commerce companies.To solve this problem,user churn prediction models were established by using statistics and machine learning algorithm in this paper.Then,the number of e-commerce enterprises were predicted and corresponding retention strategies were proposed according to the prediction results.Prediction of user churn is equivalent to classification in machine learning.Based on user historical consumption data on Jingdong mall APP,32 characteristic variables were selected from 4 dimensions in this paper,which consisted of user attributes,user consumption behavior,user monetary sensitivity and user platform dependency.Moreover,data sets were built to predict user churn.Firstly,the original data set was preprocessed by applying statistical methods in this paper,including random under sampling method,balanced data set,missing value processing and data standardization.Then the knowledge of statistics were combined with machine learning theory.In order to forecast whether users would appear loss trend,a series of algorithms were used to analyze the purchasing behavior index of e-commerce,which contained the decision tree classification algorithm in statistical classification prediction,the Adaboost algorithm in ensemble learning,random forest algorithm and MISEN*algorithm which was improved on mutual information in selective ensemble learning.Finally,the comparison and analysis of these algorithms showed that the model established by the MISEN*algorithm performed better in predicting the loss of e-commerce users.Therefore,MISEN*algorithm can be used to predict whether every user will lose or not in practical application of e-commerce user churn prediction.Then we can draw up appropriate drainage strategies based on prediction results,reducing user turnover rate as well as business losses of e-commerce enterprises.