基于社交与消费数据的反欺诈分析和建模
发布时间：2019-02-15 13:03
【摘要】：随着互联网金融及互联网大数据在国内迅速地发展,以及互联网+时代的到来,兴起了一大批的互联网金融网贷企业,与此同时用户可能发生的欺诈风险成为互联网金融的一个重要关注方面,如何有效地预见和识别潜在的用户欺诈行为,成为目前互联网金融的重要目标之一。本文试图根据某公司提供的关于社交方面和消费方面的数据,探索一个能够比较准确预测用户欺诈行为的风控模型,达到减少因欺诈带来的损失的目的。本文首先对数据进行观察和清洗,然后把清洗后的数据进行横向合并;其次,为了了解消费行为与欺诈的关系以及社交网络与欺诈的关系,做了一些探索性的描述性统计分析,并且分析了消费类型与用户欺诈关系,消费金额与用户欺诈关系以及欺诈用户紧密程度的影响,然后通过探索的结果选择合适的变量;再结合目前数据挖掘的机器学习算法和现代化金融理论对上述变量训练模型;再对模型进行对比并且对模型结果进行评价,根据评价结果再优化我们的模型;最终通过ROC曲线得出较准确的风控模型。
[Abstract]:With the rapid development of Internet finance and Internet big data in China, and the arrival of Internet era, a large number of Internet finance and net loan enterprises have emerged. At the same time, the possible fraud risk of users has become an important concern in Internet finance. How to effectively foresee and identify potential user fraud has become one of the important targets of Internet finance. This paper attempts to explore a risk control model that can accurately predict user fraud based on the data on social and consumer aspects provided by a company in order to reduce the losses caused by fraud. Firstly, the data are observed and cleaned, and then the data after cleaning are combined horizontally. Secondly, in order to understand the relationship between consumer behavior and fraud and the relationship between social network and fraud, this paper makes some exploratory descriptive statistical analysis, and analyzes the relationship between consumption type and user fraud. The relationship between consumption amount and user fraud and the influence of the closeness of fraudulent users, and then the appropriate variables are selected through the results of exploration. Then combined with the current data mining machine learning algorithm and modern financial theory to the above variables training model, then compare the model and evaluate the results of the model, and then optimize our model according to the evaluation results. Finally, a more accurate risk control model is obtained by ROC curve.
【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】：D924.35;F724.6;F832.4
本文编号：2423385
[Abstract]:With the rapid development of Internet finance and Internet big data in China, and the arrival of Internet era, a large number of Internet finance and net loan enterprises have emerged. At the same time, the possible fraud risk of users has become an important concern in Internet finance. How to effectively foresee and identify potential user fraud has become one of the important targets of Internet finance. This paper attempts to explore a risk control model that can accurately predict user fraud based on the data on social and consumer aspects provided by a company in order to reduce the losses caused by fraud. Firstly, the data are observed and cleaned, and then the data after cleaning are combined horizontally. Secondly, in order to understand the relationship between consumer behavior and fraud and the relationship between social network and fraud, this paper makes some exploratory descriptive statistical analysis, and analyzes the relationship between consumption type and user fraud. The relationship between consumption amount and user fraud and the influence of the closeness of fraudulent users, and then the appropriate variables are selected through the results of exploration. Then combined with the current data mining machine learning algorithm and modern financial theory to the above variables training model, then compare the model and evaluate the results of the model, and then optimize our model according to the evaluation results. Finally, a more accurate risk control model is obtained by ROC curve.
【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】：D924.35;F724.6;F832.4
【参考文献】
相关期刊论文 前8条
1 孙淑萍;;互联网金融风险控制[J];金融理论与实践;2014年11期
2 高汉;;互联网金融的发展及其法制监管[J];中州学刊;2014年02期
3 宗潇泳;;互联网金融相关理论问题探析[J];经营管理者;2014年05期
4 刘英;罗明雄;;互联网金融模式及风险监管思考[J];中国市场;2013年43期
5 谢平;邹传伟;;互联网金融模式研究[J];金融研究;2012年12期
6 伍永锋;;基于模糊支持向量机的电子商务交易安全风险评估方法[J];科技通报;2012年09期
7 姚潇;余乐安;;模糊近似支持向量机模型及其在信用风险评估中的应用[J];系统工程理论与实践;2012年03期
8 胡海青;张琅;张道宏;陈亮;;基于支持向量机的供应链金融信用风险评估研究[J];软科学;2011年05期
，本文编号：2423385
本文链接：https://www.wllwen.com/jingjilunwen/guojimaoyilunwen/2423385.html
