信用卡用户分类—违约预测与分析

发布时间：2018-01-30 01:32

本文关键词： 信用卡分类变量选择预测　出处：《兰州大学》2017年硕士论文　论文类型：学位论文

【摘要】：因信用卡使用给人们带来便利,越来越多的人使用信用卡。信用卡要求人们按时还款,若未按时还款,银行将收取一定的利息获益,因此各大银行为了抢占市场,最大化发展客户。但有一些客户可能会违约使银行利益受损,故控制违约情况显得迫在眉睫。本文通过一组台湾某银行的客户基本信息及违约数据,来建立模型,运用各种分类方法预测客户是否即将违约,从而对客户采取一定的限制行为,减少损失。其中分别使用以下分类方法:朴素贝叶斯、K近邻、加权k近邻、支持向量机、bagging、AdaBoost、决策树、随机森林、神经网络。比较各种方法结果,甄选出最适宜于分析信用卡数据的分类模型。其次,对数据预测中使用的变量进行筛选,筛选出重要的变量以减轻模型的复杂度,以此减少我们监督客户是否违约的成本。
[Abstract]:Because of the convenience of credit card use, more and more people use credit card. Credit card requires people to repay on time. If they fail to repay on time, the bank will charge a certain amount of interest to benefit. Therefore, in order to seize the market and maximize the development of customers, some customers may default to the detriment of the bank. Therefore, it is urgent to control the default situation. Through a group of basic customer information and default data of a certain bank in Taiwan, this paper establishes a model and uses various classification methods to predict whether the customer is about to default. The following classification methods are used respectively: naive Bayes K-nearest neighbor, weighted k-nearest neighbor, support vector machine bagging. Ada boost, decision tree, random forest, neural network. Compared with the results of various methods, select the most suitable for the analysis of credit card data classification model. Secondly. In order to reduce the complexity of the model, the variables used in the data prediction are filtered out to reduce the cost of monitoring whether the customer is in default or not.
【学位授予单位】：兰州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F224;F832.4

【相似文献】