Research On Customer Churn Modeling Methods Based On The Bias-Ensemble Learning

Nowadays with the intense competitions in the car market, it meets more challenges for the traditional marketing mode. In such case, people pay much more attention on the marketing management strategies about how to make full use of business intelligence technology for large data integration, reconstruct the marketing mode according to marketing scheme, offer multidimensional business data analysis for enterprise management, and provide information for the decision making process by data mining techniques. This paper applies diverse data mining tools to predict such customer churn behaviors, which are typical applications of customer relations management in the business intelligence. It also tries to accurately analyze several concrete issues, such as the enterprise customer analysis and marketing prediction. Therefore, this paper has very important market value, and at the same time for the method research of data mining technology has some theoretical value.In essence, the customer churn problem can be treated as an application of the classification and prediction in data mining. The real car sales and maintenance records in an enterprise are used in the paper. The data is firstly preprocessed based on the commercial facts, then a biased ensemble training method is proposed. The idea underlying such application is that the considered data behaves imbalanced. The utility of the new method is tested by comparing other methods on25groups of UCI datasets. Then, a data structure model is designed with two layers especially for the classification of three labeled data and successfully applies the biased ensemble method on the model. The model trains the local customers'information from2005to2008, and predicts the loss of customers in2009.The biased ensemble classifier uses the decision tree as base classifier on the first layer, and distinguishes the customers who are not lost from others. It then uses support vector machine for the classification of confirmedly lost and dubiously lost customers. Finally, several parameters of the model including conformed loss factor and loss factor are assessed by the mixed matrix. It is also verified that the model, which integrates the biased ensemble method and the two layered structure, shows remarkable application effects by the comparison of other classification algorithms with one layer or two layers.In general, this paper introduces a car customer churn warning model integrating the theory of data mining and practical application, which is significantly important for the analysis of enterprise customer loss. The method is also recommended applied to other areas. Besides, it also theoretically provides an alternative approach in the investigation of biased ensemble algorithm and two layered classification structure for the imbalanced learning.