碩士 / 逢甲大學 / 金融碩士在職學位學程 / 107 / This thesis conducts a predictive analysis of credit card default risk. The data are 75,109 samples of credit card holders from a commercial bank in Taiwan. The data items include the holder’s age, education, gender, marriage status, job category, line of credit, annual income, JCIC credit score and default record. The default record indicates whether the holder has been in default or no default on the payment. Decision Classification Tree and Logistic regression are used to construct the prediction models. The dataset is randomly divided into training data and testing data, 70% and 30% of the whole samples respectively. For the Decision Classification Tree model, the accuracy rates of total prediction, including default and no default, are 96.87% and 96.81% for the training and testing data, respectively while for the logistic model, the accuracy rates are 88.36% and 88.64%. For the Decision Classification Tree model, the accuracy rates of default prediction are 40.82% and 44.20% for the training and testing data, respectively while for the logistic model, the accuracy rates are 93.57% and 94.34%. Because default prediction is more important in credit risk management, thus, the logistic model outperforms the he Decision Classification Tree model. In the logistic model, the results show that age, education, gender, marriage status, annual income and JCIC credit score are significantly affect the default probability. JCIC credit score has great impacts, especially.
| Identifer | oai:union.ndltd.org:TW/107FCU01667002 |
| Date | January 2019 |
| Creators | 林佩樺 |
| Contributors | 呂瑞秋 |
| Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
| Language | zh-TW |
| Detected Language | English |
| Type | 學位論文 ; thesis |
| Format | 33 |
Page generated in 0.0015 seconds