Research For The Customer Classification Mode Of Commercial Banks Based On The Bayesian Classifier

Customer classification is the base of the implement of CRM in commercial banks. It's the precondition to analyze and forecast customer's pattern of consumption, and the premise of personalized marketing services and management. There is a large number of valuable customer information in huge amounts of data accumulated by commercial banks, which is used to identify customers and provide decision support. Data Mining (DM) technology came into being with the challenge of "rich data without knowledge". Among them, Bayesian network has been widely used in classification, with its various advantages, such as based on sound theory, expression of qualitative and quantitative, the ability of expression in probability, and the characteristics of learning with prior knowledge included. Bayesian network classifier is showing great scalability and robustness.Based on the description of customer classification models and the theory of Bayesian, this paper explored card customer classification models of commercial banks based on the Bayesian network classifier. The main work of the dissertation is as follows:(1) an overview of commercial banking customer classification model and the Bayesian network classification model; then, pointed out that there were no more literatures of Bayesian network used in the commercial banks;(2) Studied the theory of customer classification of commercial banks, and pointed out the importance of research on customer and customer classification; given the theory of customer classification of commercial banks;(3) Studied the methods of customer classification; mainly introduced decision trees, neural networks and Bayesian Network Classifiers and compared their advantages and disadvantages;(4)put forward the issue of cards customer classification; built bank card customer classification models based on Naive Bayes classifier and TAN classifier; compared the classification accuracy of Bayesian classifier with SOM neural network; analyzed the results of classification.