Classification And Regression Trees And Their Personal Credit Rating

Personal credit scoring is an important risk-defending tool in the consumer credit for the financial institutions. As a nonparametric pattern recognition skill based on statistic theory and computer science, Classification And Regression Tree(CART)has been applied widely in personal credit scoring field. The focus of this thesis is to improve the traditional personal credit scoring model by using CART. There are two main improvements in this thesis. One tries to use the variables selected by CART to build a comprehensive scoring model. The other is to use the correct sorting rate by CART and with traditional personal credit scoring methods to construct an index framework of personal credit scoring.CART can't bring a score or a default probability to clients during personal credit scoring. To solve this problem, we first combine it with linear discriminate analysis to build a model. Then we make Box-Tidwell Transform on the data and build another model. The latter model performs better on sorting validity and stability. Index system of personal credit scoring based on CART is to quantify the influence that each index performs on the consumer's credit status. Different weights are given to these indices according to the influence values. Then, the scores are calculated by using a risk measurement. The results show that the new index system performs well on evaluating the risk of the borrowers.