Research On Customer Churn Prediction And Recommendation Method Based On XGBoost And BP Neural Network

With the development of social economy,various organizations or enterprises pay more and more attention to customer relationship management,while members are customers who have high viscosities and can make greater profits contribution to the organizations or enterprises.However,for various reasons,the phenomenon of member churn has a greater impact on the operation of organizations or enterprises.By using the members' information and members' behavior information data stored by enterprises and the data mining method to predict potential churn members,and thus through collaborative filtering algorithm to recommend specific products and services in advance to retain members,which is essential for the long-term and stable development of the organizations or enterprises.The main content of this paper is to build a combined model using XGBoost(Extreme Gradient Boosting)algorithm and BP(error Backpropagation)neural network algorithm based on the membership information of an online music website KKBox,and to predict potential churn members by using the combined model.Then,the collaborative filtering algorithm is used to recommend music to the members who are predicted to be lost,in order to retain the members.The main work of this research paper includes the following three parts:First,the preparation of the collected data before modeling is carried out,which includes data cleaning,data integration,data filtering,data changing and data normalization.Then,on the basis of previous research on customer churn prediction,the advantages and disadvantages of different algorithm models are studied by analysis and comparison.Second,by using XGBoost algorithm and BP neural network algorithm,a new combined model is proposed.And thus the combined model is built and optimized,which is not only better than the single model in accuracy,but also has a great improvement in other aspects.Last,according to the user's historical information,an algorithm based on collaborative filtering is used to recommend suitable music to members who are predicted to be lost.In order to reduce the loss of members.