Research On Credit Risk Management Model And Technology Of Credit Card Based On Support Vector Machines

Credit card with consumer credit as the main function has become a very popular individual financial tool and credit cards have become an important source of income for banks. Commercial banks in the developed countries have taken credit card as an important financial product for improving the brand, attracting personal client and increasing profit. Card issuance will keep increasing with a high speed in the future years. But risk is always a companion of profit. The credit card issuers in our country are facing a challenge to their profits. The research on credit risk management of credit cards is significant for the sound development of credit card business, steady and efficient operation of the finance system, and achieving sustainable and steady economic development as well as avoiding the loss of card holders and card issuers and realizing the profits of card issuers in China.This dissertation builds up models in two stages of credit card business: one is for card application, examination and issuance, and the other is for client relation maintenance and management. In the card application, examination and issuance stage, credit scoring model based on support vector machines are constructed, and a reject inference model is built by combining two data mining techniques. In the client relation maintenance and management stage, a behavioral scoring and fraud detection model is constructed according to the consumer behavior of the client.SVM is a new technique in the field of data mining and widely applied in dealing with classification and regression. There are three main problems we may encounter when applying SVM to treating classification and regression, specifically, 1) selecting the optimal input feature; 2) the choice of kernels; 3) the determination of the kernel's parameter. In this dissertation, CART (Classification and Regression Tree) and MARS (Multivariate Adaptive Regression Splines) are used to select the input feature, and grid search to optimize model parameters. Real world date sets are used in the experiment of the proposed hybrid model.Reject inference is a term that distinguishes attempts to correct models in view of the characteristics of rejected applicants. The main difficulty in establishing a reject inference model is that the‘through-the-door'applicant population is unavailable. In this dissertation, we propose a hybrid data mining technique for reject inference. It is a three-stage approach: k-means cluster, support vector machines classification and computation of feature importance. By combining the samples of the accepted applicants and the new applicants, we obtain representative samples. To some extent, this is cost-free. Analytic results demonstrate that our method improves the predictive performance while still retaining interpretabilityBehavioral scoring is a way of updating the assessment of consumer credit risk in the light of current and most recent performances of a consumer. It is necessary to assess the credit risk by behavioral scoring due to the dynamic change of the personal credit status and the purpose of scoring of a bank. Lenders can make better decisions such as what credit limit to assign, whether to market new products to a particular client, and how to manage the recovery of the debt if the account turns bad. The self-organized neural networks and multi-class support vector machines are combined as a SOM-MSVM behavioral scoring model. The recent behavioral of a client are clustered, each cluster has similar behavioral and scoring; and each cluster is labeled with a cluster label. The classifier is set up based on those clusters. In this dissertation, SOM is used to cluster the client's consumer data, MSVM is used to score the client's behavior.Due to the dramatic increase of fraud which results in loss of billions of dollars worldwide each year, this dissertation discuss the fraud problem. The concept of fraud, the main ways of fraud, the prevention of fraud and the fraud detection model are summarized. The one class support vector machines based fraud detection model is proposed.