应用决策树评估信用卡发放风险.docVIP

应用决策树评估信用卡发放风险 摘 要 中国的信用卡行业经过 20余年的发展逐步迈入了迅速发展的快车道。然而 随着发卡量和业务量的不断扩大,发行信用卡的银行正在面临着与日俱增的信 用风险。在欧美等发达国家,有一套完整的信用评价体系来控制信用卡的信用 风险,但中国目前还缺少这样的个人信用体系。因此如何有效控制信用卡信用 风险成为中国信用卡行业普遍关注的问题。 评估信用卡发放风险是控制信用卡信用风险的第一道防线。在信用卡发放 过程中能否有效地识别、拒绝那些高风险的申请者,将在很大程度上决定发卡 银行的信用卡资产质量和坏账风险。如果能够在信用卡发放前有效识别出可能 给银行带来信用风险的客户,将会给接下来的风险控制工作带来很大便利。 评估信用卡发放风险的方法有经验判断法、层次分析法、传统统计线性模 型以及数据挖掘法。本文通过应用数据挖掘中的 CART决策树算法,在台湾某 商业银行信用卡的实际数据集上进行实证分析,识别信用卡发放过程中可能给 银行带来信用风险的“坏用户”。本文一共提出了七个模型:模型一、二有比较 好的分类准确率,但是存在将过多坏用户误判为好用户的问题。模型三、四通 过调整好客户和坏客户的抽样概率以及引入损失函数两种方法很好地解决了这 一问题。模型五提出了只利用申请者个人基本信息建立的决策树模型。模型六 将好用户根据其信用卡使用频率和月刷卡额进一步划分为四类,然后对这四类 具有不同客户价值的申请者进行判别。模型七通过引入 Bagging 算法解决了模 型五的不稳定性,并提高了预测精度。 总的来说,本文深入研究了 CART 决策树算法在评估信用卡发放风险中的 应用,取得了不错的效果,为我国控制信用卡信用风险提出了一条行之有效的 途径。关键词:信用卡发放;风险评估;数据挖掘;CART决策树Abstract After 20 years of development, Chinas credit card industry is gradually entering an express way. However, with the growing number of cards and expanding volume of business, the banks who issue credit cards are facing increasing credit riskIn Europe and the United States, commercial banks have a complete credit evaluation system to control the credit risk of credit cards, but China still lack such a system. Therefore, how to effectively control the credit risk of credit cards is becoming issues of common concernAssessment of risk in issuing credit cards is the first line of defense in controlling credit risk of credit cards. Whether or not high risk applicants can be effectively identified and refused in credit card issuance process, will largely determine the asset quality and bad debt risk of card issuers. If the customers who may bring card issuers credit risk can be recognized effectively, it will give convenience to the next stage of risk controlThe methods that can assess the risk in card issuance process include empirical judgments method, hierarchical analysis, the traditional statistical linear model and data mining. In this thesis, the CART decision tree algorithm