Research On Customer Churn Early-warning Based On IG_NN Double Attribute Selection

In the face of the fierce market competition and changing market demand, it is the parent of success for every enterprise to reduce customer churn to the maximum extent.As a very important part of customer relationship management theory, the core of customer churn management lies in how to conduct effective customer churn early-warning. At present, the technology of data mining is widely used in the study of customer churn early-warning.In the background of customer relationship management, the paper discusses the problem of customer churn early-warning deeply by using data mining, based on a comprehensive summary of customer churn management. To begin with, through extensive practice of customer churn management, the concept, basic characteristics and functions of customer churn early-warning are studied to research the system of it. In the next place, to eliminate the subjectivity of attribute selection and improve accuracy rate, coverage rate, hit rate and lift coefficient of model to avoid serious overlap of different types of customer attributes in attribute space,the paper proposes the customer churn early-warning model based on double attribute selection of information gain(IG) and neural network(NN) by analyzing the characteristics of customer churn data and the particularity of customer churn early-warning(large volume of data, high dimension of data, unbalanced data set, attribute selection is often subjective and speculative, etc.),in order to improve prediction performance of model and enhance the accuracy rate of prediction. That is,firstly, undertake the main attribute selection for customer churn data by using IG, and then analyze every main attribute by using NN, which output results are analyzed by 80-20 rule to get the key attributes affecting customer churn; secondly, construct the customer churn early-warning model based on IG_NN by taking the key attributes as input and customer churn probability as output.To verify the validity of method, customer churn early-warning model based on IG_NN double attribute selection is applied to customer churn early-warning of a telecom operator in Inner Mongolia.Meanwhile,the model is compared with single early-warning model such as decision tree and neural network. Provenly, at the current time conditions,there is improvement of various degree on accuracy rate, coverage rate,hit rate and lift coefficient than other methods for customer churn early-warning by the model.Thus,The model has a good prediction performance for dealing with a large quantity of non-equilibrium data set. The paper, in theory, improves the basic theories and ideas of customer relationship management and enterprise early-warning; in practice, it can help enterprises avoid the risk of customer churn,based on a good prediction performance of the model, and effectively guide them to implement customer relationship management to enhance their competitiveness.