Research On Application Of Data Mining Technology In Communication User Churn Warning

Customer churn management plays a key role in customer relationship management.It can predict potential lost users by building a user churn early warning model,thus it can assist enterprises to make early warning and take corresponding retention measures in time.Communications operating enterprises have a huge number of customers and master a huge amount of user data.In today's saturated communication market,operating enterprises need to fully explore the value of data and strengthen the management of stock users.High-value users play a vital role in enterprise operation.In order to further reduce the maintenance cost of the enterprise,it is necessary to further identify highvalue users in the potential customer loss group.The application of data mining technology in the prediction of the loss of communication users can quickly and accurately classify the communication users,so as to achieve the purpose of improving the success rate of enterprise retention,reducing the operating cost and promoting the operating profit of enterprises.The purpose of this paper is to apply data mining technology to the prediction of user churn in the communication industry,and integrate the underlying models to improve the prediction accuracy of the model.On this basis,it can further identify highvalue users as the retention customer group and thus save the maintenance cost.This paper uses mobile users in a province as the research object.Random Forest,XGBoost,Logistic Regression and Naively Bayes algorithms were used to build the churn warning models respectively,and the model parameters were adjusted.On this basis,Voting fusion methods(including Hard Voting and Soft Voting)were used to fuse the four models in different combinations.Based on the research,this paper mainly draws the following conclusions: first,the lost users usually have the following service characteristics before they leave the network: the age group is in a low range,the user stay on the network for a short time,there is the over-set behavior,the ARPU level is low,and the flow fluctuates greatly;Second,in a single model,the model built by XGBoost algorithm has the best prediction effect,with the prediction accuracy reaching 90.01%,while the model built by Naive Bayes algorithm has the worst prediction effect,with the prediction accuracy only reaching 78.42%.Thirdly,in the process of the fusion of the underlying models,this paper explored the optimal combination of the underlying models under the Voting fusion method,and finally found that the fusion model obtained by the fusion of random forest,XGBoost and Logistic Regression models in the Soft Voting method had the highest prediction accuracy,reaching 92.82%.Finally,this paper uses the churn warning model to forecast the churn of mobile users in a province in November 2020,and the total number of churn users in November is 278,951.And on this basis to introduce the concept of customer value,that is in the normal network users,the consumption amount of consumers in the last three months is at least 50 yuan or the top 30%.Among the potential lost users output by the model,those who meet the high value condition are regarded as the customer group to maintain of the mobile enterprise,and the customer group of 69,737 is finally obtained,further narrowed the scope of customers to maintain.This paper not only enriches the research of fusion model,but also puts forward the concept of user value based on the construction of churn warning model,and further refines customer retention,so as to achieve the purpose of obtaining high revenue with low operating cost.