MARS-based Research of Personal Credit Scoring: Verification of Chinese Data

In order to fulfill the commitment that the Chinese government promised when China joined the WTO, the banking industry of China will allow foreign capital to begin having a share in all banking business at the end of 2006. It is significant for Chinese banks to develop credit card services to compete with foreign capitals. On the other hand, it is urgent for them to improve their ability to control the credit risks. This paper introduces the principles of the multivariate adaptive regression splines (MARS) classification method, and analyzes the relationships between MARS and CART in the perspective of these principles. Also this paper contrasts the classification accuracy rate and robust attribute of both methods by applying a case of true personal credit data from a city of China. The results are that MARS model, which is based on actual Chinese data, has a good classification accuracy rate and robust attribute, and it provides an alternative so that the Chinese banking business can do personal credit risk management better