Broadcasting Customer Loss Based On Double-layer Feature Selection Predictive Model Research

With the rapid development of information network technology and the popularity of network set-top boxes,the loss of user rates for cable network operators across the country is on the rise.The competition among the three major operators is getting more and more serious,and their products and services are different from before.Diversified products and high-quality services have emerged.According to the survey,in the first half of the year,the national IPTV has occupied a market share of 31.77% per month,and is continuing to grow.It is imminent for the radio and television industry to take effective measures to retain the market.The application of data mining technology in the maintenance of customer relationships among the three major operators of telecommunications,China Unicom and China Mobile has brought good market response.However,according to previous market research,it was found that the application of data mining in the radio and television industry started late.Although radio and television operators in various regions have gradually established enterprise operation analysis systems,they mostly stay at the level of business analysis reports and do not conduct deeper data mining.Making full and reasonable use of the large amount of basic data stored by the enterprise,and applying data mining to explore the new model of radio and television customer management based on the characteristics of the radio and television industry,will bring considerable economic benefits to radio and television operators.This article first analyzes the characteristics of the radio and television industry,and points out the difficulties in using the radio and television customer data set to predict customer churn,that is,the problem of high current feature dimensions and imbalanced data samples in the radio and current customer data set.On this basis,in order to solve this problem,the paper starts from the perspective of optimizing the feature selection method of the prediction model,synthesizes the advantages and disadvantages of various required feature selection methods,and the scope of application,and points out the feature selection method based on search strategy partitioning:exhaustive search,Sequence search and random search feature selection methods are not applicable to radio and television customer churn prediction.And the feature selection method based on the evaluation criteria: the problems encountered when filtering and packaging are applied to the prediction of customer churn in radio and television.In order to obtain a better radio and television customer churn prediction model,this paper first proposes a two-layer feature selection method that combines the advantages of the filtered feature selection method and the packaged feature selection method.On the one hand,it solves the problem of excessive feature dimensions of radio and television customers;on the other hand,It can also ensure the accuracy of the model's prediction effect.The idea of constructing a radio and television customer churn prediction model based on double-layer feature selection is that the feature selection part adopts a filtered feature selection method as the first-level feature selection method,and then selects the optimal solution of the feature combination based on the comparison of the prediction effects of each classification model..Combined with a combined classifier based on decision tree algorithm,Logistic regression algorithm and neural network algorithm,the effect of radio and television customer churn prediction model is improved.In order to verify the true prediction effect of the model,this paper uses the real business data of a cable TV operator in a province and constructs a prediction model according to the data mining process standards to achieve the evaluation of the application effect of the customer churn prediction model of radio and television.The data analysis shows that the two-layer feature selection method based on Fisher's ratio and AUC index evaluation criteria can improve the prediction effect of the classifier.By combining the construction of the classifier,the generalization of the broadcast churn prediction model can be improved,and the calculation time Within the acceptable range of the enterprise.The new model can help radio and television operators maintain market share to a certain extent and improve business efficiency.