Research On Telcom Customer Churn Prediction Based On Deep Learning

With the popularity of the Internet,the communication market has become saturated with users,the cost of developing a new customer is much higher than that of retaining an old customer,and enterprises are shifting from attracting new customers with products to retaining old customers.At the same time,with the development of data mining technology and deep learning,enterprises pay more and more attention to using technology to predict the customers churn.Based on real enterprise data,this paper establishes a forecast model of telecom broadband customer churn based on deep belief network.Combining the advantages of machine learning and deep learning,eXtreme Gradient Boosting is used to create features and the deep belief network is used for modeling.Compared with deep neural network,the deep belief network has better global search ability,which makes the model converge to the global optimal solution faster and better.This paper firstly analyzes the methods of customer churn prediction domestic and abroad,and then expounds the algorithms and knowledge points involved in establishing the forecast model of telecom customer churn.Restricted boltzmann machine,Deep neural network,eXtreme Gradient Boosting.Then the basic customer information table,customer activity information table,customer complaint table and optimical power table of XX city were extracted from the system of XX provincial telecom company.Perform data integration,data cleansing,data specification,and data transformation for these tables.The extreme gradient tree is used to create new features,and then the deep belief network is used to establish the telecom customer churn prediction model.The accuracy of the model reached 90.45%,the recall rate reached 88.11% and the F1 score reached 89.27%,use the August telecom data as test data.Meanwhile,Deep neural network,Lightgbm and random forest were used for model comparison.The experiment shows that the deep belief network based on the eXtreme Gradient Boosting can deal with the problem of telecom customer churn well.