信用卡是一種具普及性的現金支付工具並且被廣泛運用,若持卡人未如期繳款可能會造成信用卡違約支付風險,故信用卡違約付款分析預測是重要議題。本研究以UCI公開信用卡資料集屬性包括銀行貸款金額、持卡人過去還款狀況、信用卡帳單金額等,進行資料探勘分析,分析各屬性資料與違約付款的關聯性,同時以隨機森林特徵排序分析重要屬性。本研究應用資料探勘方法的決策樹、隨機森林、神經網路進行建模與預測,根據實驗結果準確度皆可達百分之八十、又以神經網路效果最佳。透過隨機森林排序與資料探勘分析結果顯示教育程度、婚姻狀況、信用卡還款狀況與信用卡帳單金額是信用卡違約付款的關鍵因素。
Credit card is a popular cash payment tool and is widely used to replace the cash. If the cardholder fails to pay as scheduled, the risk of credit card late payment may be incurred. Therefore, credit card late payment forecasting is an important issue.This study analyzes UCI credit card data set, which includes attributes such as bank loan amount, cardholder's past repayment status, and credit card bill amount, etc. This study analyzes the relevance of each attribute to the late payment and discovers the important attributes by using random forest technique. This study applies data mining methods such as Decision tree, Random Forests, and Neural Networks to build the prediction model. According to the experimental results, the accuracy can reach 80%, and Neural Networks can achieve the best performance. The analysis of this study shows that the education degree, marriage status, credit card payment and credit card bill are the critical factors for predicting credit card late payment.