【作者】 庄玮；

【导师】 张育平；

【作者基本信息】 南京航空航天大学 ， 计算机应用技术， 2008， 硕士

【摘要】 随着经济和社会的发展,以及全球金融市场的不断开放,各国政府积极推动金融自由化与国际化的措施,而以信用卡作为媒介的交易行为不断激增。但是,伴随发卡量的大幅增长和交易量的不断提高,信用卡欺诈呈现快速增长的趋势,且欺诈手法不断翻新,使银行很难迅速有效的从大量交易记录中觉察出欺诈交易,由此带来了巨大的风险和损失。因此,迫切需要一个能对信用卡交易进行快速判断和准确识别的模型或系统来辅助银行的工作。本文针对我国银行信用卡交易中普遍存在的欺诈问题,依据数据挖掘技术,构建信用卡欺诈行为识别模型,为我国银行的信用卡风险管理提供技术支持。本文首先简单介绍了我国信用卡风险管理的现状,分析了欺诈风险的成因和识别防范策略。然后,运用自组织映射(SOM)神经网络算法和组合分类器原理,构建基于数据挖掘的银行信用卡欺诈识别模型:先采用SOM网络算法将数据量庞大的样本集进行初步的分类处理,以提高再次分类的准确性,然后将得到的训练子集分别与欺诈样本结合形成新的子集,接着根据客户分类指标,再次利用SOM网络算法对前面得到的各个子数据集进行分类,最后采用投票法将分类结果融合,从而建立信用卡客户分类的组合模型。在此基础上,依托中国银行下属相关支行的信用卡交易具体数据,对构建的欺诈识别模型进行分析,验证了该模型的实际运用效果。更多还原

【Abstract】 With the development of economy and society, and the continuous opening of financial markets around the world, government of each country is actively promoting kinds of measures relevant to financial liberalization and internationalization, transactions based on credit card increase unceasingly. However, with the large-scale rise of credit cards and volume of transaction , the enlargement of business scope, and the quick growing of market, the amount of credit card fraud is boosting on amazing speed, and fraud methods have been retrodden and commit skills have been shrewder day by day. It is difficult for banks to discover fraud transactions effectively, and the risk and loss is larger and larger, so a model or system which can quickly judge and accurately distinguish credit card transaction is in urgent need to assist bank work.This thesis is directed at the fraud matter existing in the bank credit card transactions of our country. It constructs the detection model of credit card fraud on the basis of data mining technology and provides technical support for risk management of our bank credit card.This thesis firstly introduces status of credit card risk management of our country and analyses the causes of fraud risk and tactics of fraud detection and prevention. Then it constructs credit card fraud detection model based on data mining technology, using Self-Organizing fenture Map(SOM) arithmetic and combined classifier theory: the large number of sample collection is tentatively classified with SOM to improve the accuracy, then, the trained collection and fraud sample is combined respectively to form new collections ,and then, classifying the preceding collections once more with SOM based on client classified quota, and finally melting together the classified results with ballot, thus the classified modular model of credit card clients is set up. Lastly, it analyses the constructed model on the basis of concrete data of credit card transactions provided by sub-branches of Bank of China, and confirms the effect of the model with living examples.更多还原