Research On Integrated Cost-sensitive Classification Methods Of Customer Churn Prediction

Customer-focused management philosophy has become the cornerstone of enterprise development, as a characteristic of the customer relationship management technology, has become an important means to gain competitive advantage. Customer churn is an important component of customer relationship management. The results show that, if effectively reduce customer churn rate, will be able to greatly enhance the competitiveness of enterprises and the level of profitability. Therefore, in an increasingly competitive global market today, find out the reason of customer churn and reduce customer churn rate is an important theoretical and practical significance of the research topic.Firstly, this paper reviews the basic concepts of customer churn, and the problem of customer churn means, while data mining analysis of the traditional classification in customer churn prediction inadequate. Then, elaborated classification of commonly used data mining, with emphasis on decision tree and support vector machine method for the improved algorithm proposed later theoretical groundwork done. Next, for the traditional classification methods in the classification process assumes that all classification errors are equal misclassification costs of the limitations; the introduction of cost sensitive learning mechanism described for the traditional classification cost sensitive learning method is introduced to two ideas. Finally, the generalization ability of the traditional classification of weak constraints, described by introducing the ensemble learning method to solve the traditional classification does not build a single classifier stability and improve the accuracy of their learning, to effectively improve the generalization of the traditional classification methods Capacity-building.Secondly, the paper based on C4.5 decision tree algorithm and support vector machine algorithm as a benchmark by introducing a cost sensitive learning, traditional classification methods were carried out on two sensitive transformation of the price, then the classical integration algorithms for both Boosting and Bagging to integrate learning, which made the decision tree based on Cost-sensitive Boosting with Bagging and the price-sensitive and based on support vector machines cost sensitive classification of the two integration methods, and in 10 UCI machine learning database of non-equilibrium binary algorithm on data set Experimental analysis, we found two algorithms relative to their baseline algorithm can achieve better prediction performance. In addition, classification is given based on data mining to build customer churn prediction model assessment index for use in case analysis.Finally, the customer data sets of Personal Financial Services of a Chongqing commercial bank as the analysis of sample data sets, using the two proposed integrated classification cost sensitive customer data sets in the churn prediction models were established, and predictions and common data Mining classification algorithm: RBF neural networks, Bayesian networks and Logistic regression prediction model established by the results of a comparative analysis of two algorithms proved a good predictor of performance. View of the model can be explanatory, the paper-based and Cost-sensitive Decision Tree Boosting Algorithm for the model is integrated as a predictive model of the case to discuss the decision of the prediction model and explain the contents of rule extraction.