Research On Personal Credit Scoring Model Based On Data Mining

With the rapid development of China economic, consumer credit becomes more and more important as personal loans grow rapidly. Personal consumer credit has been a significant factor to stimulate domestic demand and to promote economic growth. One main problem emerging in the progress of the development of consumer credit is the difficulty in assessing and controlling the individual credit risk, resulting in the huge risk of consumer credit business, so the research on personal credit scoring models is of significance in practice.The credit card is one of important financial credit payment instruments with the fastest growing speed and most innovative vitality. In this paper, we study personal credit scoring model based on credit card risk measure. We introduce data mining methods such as Support Vector Machines (SVM), Genetic Programming (GP) and Genetic Algorithm (GA) to credit card risk management, which used to build the credit risk scoring models and the behavior scoring models. Lastly, the presented models are successfully applied to the real credit data and the efficiency of the models is verified.Firstly, we develop a two-stage credit risk scoring model (2SRSM) by integrating the advantage of GP and SVM. We also present the misclassification costs and multi-objective optimization strategy to make the model satisfy the real situation. First, we use GP to derive the IF-THEN rules. Then, the training data set which does not match those IF-THEN rules or matches more than one rule is used to train SVM and build the discriminant function. Empirical study shows that this model has many advantages such as high classification accurate rate and robust explanation capability.Lastly, we analyze a real life credit data set from a major Chinese commercial bank. In China's case, insufficient samples and high dimension are the notable characteristics of Chinese credit data. This paper applies GA to select key attributes and eliminate the impact of redundant attributes. Then, we apply multi-class SVM to build the behavior scoring model, which is used to dynamically decide customer's levels, and help decision makers effectively manage their customers. Empirical study shows that this model can select the key features and obtain high classification accurate rate.This study introduces two novel data mining techniques GP and SVM to make a further development on credit scoring models, and proves the efficiency of this model by practical test.