Applying Data Mining Techniques on the research of Credit Risk - An Example of Collection Scoring Model

• Main Authors: Ming-Chiao Liang, 梁明喬
• Other Authors: Yung-Sun Lee
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/47z6zs

碩士 === 銘傳大學 === 資訊管理學系碩士在職專班 === 95 === Credit Risk becomes a major risk issue in Taiwan banking industry. In order to control this problem well, the Basel committee requires all banks to adopt the moderate risk assessment method and to establish their own credit risk evaluation model. In the past, researches regarding Credit Scoring Model mainly focus on credit card market, and to predict card holders’ behavior in terms of the probability of future contract broken. To find the best parameters of this model, we investigate risk issues through the Data Ming technology. We used Statistics analytical method and Neural Networks algorithms alternately. To set up the best Collection credit Scoring model, the research increase parameters progressively and test repeatedly in Decision Tree model. Though using these two kinds of credit risk models, banks could assess credit card applicant effectively and the holder''s credit risk which could both raise the loan and Collection efficiency and well reduce the credit risks. This results we found that , the most critical Credit Risk factors are “already overdue days in order”,”customer refund rate”, ”telephone effective rate”,“overdue payment number of times”, ”the number of debit card”. The hit rate of the prediction model reached above 75%. The stability tests of the model indicated stable errors of ± 3% for the data of two consecutive months.