|
In recent years, the bank industry actively introduce and plan to practice the Basel Ⅱ and adjust the inside system gradually. Banks except itself had consume private liability variables, also taken the relevant consume situation of credit from JCIC. Banks actively set up default risk model in order to calculate default risk’s score and then set up and comment etc. In order to distinguishes from the size of customer's risk and reduce the loss. Then how to set up appropriate default risk models is also the key of this research. The objective of this research is to find consume private liability variables and relation of individual credit variables which have significant influences on default, and to decide the suitable cut-point of probability of expect default of the consume credit loan default by using Classification and regression tree technique. Result, this research sets up six kinds of evaluation models and finally determine “relation of bad – decision tree” is the better model by comparing six prediction abilities of model (Accuracy rate, Recall rate , FP rate, TN rate, FN rate and Precision rate ), ROC curve, and Cumulative Lift curve. In “the relation of bad - decision tree” , the important eight variables are “gander”, “education levels”, “the address of company (city)”, “inquire of other banks last three months”, “difference of total credit and total loan”, “the rate of times fully pay up last N months”, “the number of credit cards could borrow money”, “the average of credit”, etc. The decision tree that had ten levels and twenty-six nodes cut-point is 23.16% with the suitable probability of expect default.
|