基于随机森林模型的P2P借款人信用评估研究

发布时间：2018-05-06 08:18

本文选题：借款人信用评估 + P2P网络借贷　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：P2P网络借贷进入中国已有10余年的时间,目前已成为个人以及中小微企业进行融资的重要渠道。2016年全年,我国网络借贷平台的累计数量已达5877家,而正常运营的平台数量却只有2448家,平台成交量却达到了20638.72亿元,行业总体贷款余额更是高达8162.24亿元。虽然正常运营赢得平台有所减少,但是相较2015年去年的成交量与贷款余额分别增长110%和100.99%,表明P2P网络借贷还是解决融资问题的一种重要方式。近年,随着我国互联网金融整改行动的逐步深入,我国网络借贷行业的外部借贷环境逐步得到改善,但P2P网络借贷的借款人信用风险问题依然是影响网络借贷行业发展的重要因素,且国内大多数P2P网络借贷平台的信用评估方式都较为简单、评估性能也不大理想,对借款人的信用管理还有待加强。因此,本文通过随机森林模型,结合翼龙贷的数据,构建了P2P借款人信用评估模型对P2P平台中借款人存在的信用风险问题进行评估。本文在研究P2P借款人信用评估过程中,首先简要介绍了P2P网络借贷的相关概念及发展现状,其次对比介绍了随机森林、支持向量机、Logistic回归模型,确定随机森林为本文P2P借款人信用评估所采用模型;最后结合翼龙贷数据,确定P2P借款人信用评估指标体系,并利用随机森林构建了P2P网络借贷的借款人信用评估模型。在构建随机森林模型过程中,首先参照其他P2P网络借贷地相关研究初步选取了影响借款人信用风险评估19项指标;其次利用随机森林特征选取方法筛选出8项指标作为本文P2P借款人信用评估指标体系;然后利用R语言分别构建了随机森林和支持向量机,并对其分类性能进行了比较分析。实证结果表明,随机森林特征选择能有效地筛选指标,且发现在P2P网络借贷的信用评估方面,随机森林模型具有更好的分类性能。通过分析指标筛选结果与实证结果还发现,借款人在平台中的里斯借款记录及借款信息都是影响P2P网络借贷借款人信用风险的重要观测指标。
[Abstract]:P2P network lending has been in China for more than 10 years and has become an important channel for individuals and small and medium-sized enterprises to raise funds. In all of 2016, the total number of online lending platforms in China has reached 5877. But the normal operating platform number is only 2448, the platform turnover has reached 2.063872 trillion yuan, the industry overall loan balance is as high as 816.224 billion yuan. Although the normal operation won platform has been reduced, the volume of transactions and loan balance rose 110 percent and 100.99 percent, respectively, compared with last year in 2015, indicating that P2P network lending is also an important way to solve the financing problem. In recent years, with the gradual deepening of China's Internet financial reform, the external lending environment of China's network lending industry has been gradually improved. However, the credit risk of the borrowers of P2P network lending is still an important factor affecting the development of the network lending industry, and most of the domestic P2P network lending platform credit evaluation methods are relatively simple, the evaluation performance is not ideal. The credit management of borrowers needs to be strengthened. Therefore, based on the random forest model and pterosaurus data, this paper constructs a P2P borrower credit evaluation model to evaluate the credit risk of the borrower in P2P platform. In the course of studying the credit evaluation of P2P borrowers, this paper firstly briefly introduces the related concepts and development status of P2P network lending, and then compares the stochastic forest, support vector machine and logistic regression model. The random forest is the model used in the credit evaluation of P2P borrowers. Finally, the credit evaluation index system of P2P borrowers is established with pterosaurus loan data, and the credit evaluation model of P2P borrowers is constructed by using random forests. In the process of constructing a stochastic forest model, 19 indexes of credit risk assessment of borrowers are preliminarily selected according to the relevant studies of other P2P network lending sites. Secondly, 8 indexes are selected by the random forest feature selection method as the credit evaluation index system of P2P borrowers, and then the random forest and support vector machine are constructed by using R language, and their classification performance is compared and analyzed. The empirical results show that the stochastic forest feature selection can effectively screen the index, and it is found that the stochastic forest model has better classification performance in the credit evaluation of P2P network lending. Through the analysis of the index screening results and the empirical results, it is also found that the loan records and loan information of the borrowers in the platform are important observation indicators that affect the credit risk of P2P network borrowers.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.4;F724.6

【参考文献】