决策树分析技术在保险企业反洗钱监控系统的应用研究

发布时间：2018-07-06 18:01

本文选题：大额和可疑交易 + CHAID算法　；参考：《湖南大学》2014年硕士论文

【摘要】：随着金融业的蓬勃发展，我国金融业面临越来越多的安全风险。其中，，将非法收入以各种形式注入金融体系的洗钱犯罪活动也在快速增长，给国家和社会带来了巨大危害。为了规避此类风险，确保我国金融业健康持续发展，建立健全的中国反洗钱监控体系也显得日益重要。从目前掌握的案例来看，洗钱犯罪活动主要表现为犯罪分子将非法收入通过银行和保险公司等金融机构，并利用其产品的复杂性将其转化为合法收入。目前，我国保险业反洗钱监控手段停留在依据法律规定设定筛选条件，但是保险企业自身对大额和可疑交易的筛选并无创新。本文着眼于我国保险业反洗钱监控的应用研究，根据既定的法律法规及保险公司的反洗钱经验数据，在现有的数据挖掘技术研究基础上，重点对决策树主要算法展开研究，本论文的主要成果概括如下：利用决策树CHAID算法、CART算法和QUEST算法对大额和可疑交易的相关关系进行分析，以获得影响大额和可以交易识别结果的因素。然后，在此基础上，通过设立错分成本对模型进行进一步分析，使得模型结果拟合度提升，最终根据模型结果使得保险企业能获得更加具体的筛选条件。实验数据表明，CHAID算法、CART算法和QUEST算法的模型拟合度基本持平，但是在大额和可疑交易识别的准确率上均未能达到理想水平。因此，本文在原有基础上进一步优化模型，设立错分成本，重新使用CHAID算法拟合模型，准确率大幅提高，具体表现为将大额和可疑交易的对象错分为良好信用客户的概率低于15%。在大额和可疑交易的筛选自变量范围中添加投保与退保天数间隔及缴费模式，并将之应用到现行保险业反洗钱监控系统中。分析了两种筛选情况的结果差异，证明了改进后的筛选条件能提高筛选的覆盖率和准确率，最后以此给出了对于保险业反洗钱系统政策上和技术上的建议,具有一定的现实意义。
[Abstract]:With the vigorous development of financial industry, China's financial industry is facing more and more security risks. Among them, the crime of injecting illegal income into the financial system in various forms is also growing rapidly, which brings great harm to the country and society. In order to avoid such risks and ensure the healthy and sustainable development of China's financial industry, it is increasingly important to establish a sound anti-money laundering monitoring system in China. According to the current cases, the criminal activities of money laundering mainly show that the criminals turn the illegal income through banks and insurance companies, and use the complexity of their products to convert it into legal income. At present, China's insurance industry anti-money laundering monitoring means stay in accordance with the provisions of the law to set screening conditions, but the insurance companies themselves to large and suspicious transactions screening has not been innovative. This paper focuses on the application research of anti-money laundering monitoring in insurance industry in China. According to the established laws and regulations and the anti-money laundering experience data of insurance companies, and on the basis of the existing data mining technology, this paper focuses on the research on the main algorithms of decision tree. The main achievements of this paper are summarized as follows: the correlation between large amount and suspicious transactions is analyzed by using decision tree algorithm cart and request algorithm to obtain the factors that affect the results of large amount and transaction identification. On this basis, the model is further analyzed by setting up a misdivisional cost, so that the model result fit is improved, and finally the insurance company can obtain more specific screening conditions according to the model results. The experimental data show that the model fitting degree of cart algorithm and quest algorithm is basically the same, but the accuracy of large amount and suspicious transaction recognition is not up to the ideal level. Therefore, this paper further optimizes the model on the basis of the original, sets up the misdivision cost, reuses the Chaid algorithm to fit the model, and the accuracy is greatly improved. The concrete performance is that the probability of classifying the objects of large amount and suspicious transaction into good credit customers is lower than 15%. In the range of screening independent variables of large and suspicious transactions, the interval of days between insurance and withdrawal and the mode of payment are added, and applied to the current insurance anti-money laundering monitoring system. The differences between the two screening conditions are analyzed, and it is proved that the improved screening conditions can improve the coverage and accuracy of the screening. Finally, the policy and technical suggestions for the insurance anti-money laundering system are given. Has certain realistic significance.
【学位授予单位】：湖南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP277;F842.3

【参考文献】