Research On Credit Behavior Evaluation Model Based On GA And M-SVM

The rapid development of Chinese economic stimulates the enormous growth of credit consumption and the dramatically warming up of various consumer loans. Credit card, as a high-yield high-risk financial product, is the fastest growing and most innovatively vigorous financial credit payment instruments in credit consumption and has become an important object the domestic and foreign financial banks fiercely compete for. The management levels of financial banks on credit card business directly affect its profits, therefore, how to establish an effective credit evaluation model to analyze the historical data of credit card customers and consuming behavior as well as to strengthen the management and maintenance of customer relationships, is the hot subject to which the financial banks currently pay extensive attention. This subject not only has a certain kind of theoretical research meaning but also pays much importance on financial banks developments and profits realization.Based on the existing research, this paper builds a credit behavior evaluation model based on genetic algorithm (GA) and multi-class support vector machines (M-SVM) named GMBSM. It will be provide some decision such as adjustment of credit rating and credit limit for financial institutions. The outline contents in this paper describes as follows:Firstly, this paper builds a framework of behavior credit evaluation model based on the existing research. It consists of three parts:Data source, credit evaluation index system and core algorithm. Data source is including four parts:the applicant information databases, credit card accounts database, credit card transaction databases, credit card fraud databases. The basis of the model is credit evaluation index system. High-performance algorithm is the core of the model.Secondly, after studying the credit evaluation index system both at home and abroad, from the perspectives of the credit risk and customer value to consider the behavior of the cardholder's. Then build card credit evaluation index system based on consumer behavior in the guide of the principles about credit evaluation index system.Thirdly, there are the notable characteristics of the credit behavioral data—small data sample, high dimension and quickly update. This paper applies genetic algorithm to select key attributes and eliminate the impact of redundant attributes. Then propose two algorithms:BTMSVM and BTMISVM. BTMSVM algorithm is based on multi-class SVM and Binary. BTMISVM is a multi-classification algorithm which support for incremental learning.Fourthly, do experiments and analysis of the above model and algorithm. The results show that the model is effective, and has a higher classification accuracy rate. It has found that the model provide an important decision of adjust the customer credit rating and credit limit.