【作者】 汪洋；

【导师】 吴清烈；

【作者基本信息】 东南大学 ， 管理科学与工程， 2018， 硕士

【摘要】 大数据时代下,随着互联网金融行业的快速成长,网络金融服务渠道逐渐丰富,金融服务也逐渐多元化,P2P网贷平台运营面临的环境也越来越复杂。P2P网络借贷平台对个人用户行为模式的认识、金融风险安全的认识、经济发展新趋势的认识等等均需要重新确立。在此情况下,网贷平台面临的风险日趋复杂化,多样化,针对网络借贷平台的个人信用风险控制模型和体系亟待重构和优化。本文将针对P2P网贷平台为应用对象,研究基于大数据技术的信用风险控制模型。本文首先分析讨论了大数据背景下P2P网贷平台发展现状与面临的主要挑战,并分别从传统风险控制的现状和方法两个维度去分析其所面临的困境。研究分析了互联网金融迅速发展带来的金融数据多样性问题,以及它对传统风控模型的冲击;并从信息不对称理论出发,对平台借贷匹配率问题给出了理论解释。对于现有的P2P信用风控模型,本文重点分析了讨论了其设计客观性问题和隐含信息发掘问题,从而引出本文针对以上问题的解决方法研究。对于第一类的现状问题,即数据多样性和信息不对称引发的分析困境,本文提出了一种基于大数据分析的P2P网贷平台信用评估模型,在传统信用指标基础上,加入了对用户个性与品质的评估,即软信息,使得评估模型更加全面完善。本文利用数据挖掘方法整合更全面的用户信息,并通过随机森林算法计算筛选重要评价指标。基于此模型,本文计算验证了本文的风险评估模型的可行性和有效性。对于第二类问题,即传统分析模型面临的客观性和隐藏关联发现问题,本文在大数据研究基础上,进一步提出了一种基于深度学习技术的信用风险控制分析模型。本文基于卷积神经网络算法,以时间序列组织数据,构建出一套深度学习的信用评估模型。为了验证模型的性能,本文将训练优化后的模型与逻辑回归、决策树和随机森林模型进行对比,以ROC曲线和AUC数值作为评价指标。结果表明深度学习模型具有更高准确度,这也意味着深度学习模型在分析复杂非线性且具有多特征对象时具有更明显的优势。更多还原

【Abstract】 In the era of big data,with the rapid growth of the Internet finance industry,numerous online financial service channels has occured and the financial services have become increasingly diverse.As a result,the manaagment of P2 P online lending platforms faces greater challenges than ever before.The understanding of individual behavior patterns,financial risks and the new trends in economic development of the P2 P online lending platform demands to be re-established.Under such circumstances,the credit risk control model and system for the internet lending platform need to be reconstructed and optimized,as the risks faced by the online lending platform are becoming increasingly complex and diversified.Therefore,in this study,we further analyze the credit risk control model based on the big data technology,which is applied for P2 P online lending platform.First,the current situation and major challenges faced by the P2 P online lending platform in the context of big data has been analyzed,and then the difficulties from the perspectives of current sutuation and methods of traditional risk control has also been studied respectively.Meanwhile,the diversity of financial data brought about by the rapid development of Internet finance and its impact on traditional risk control models have been studied.Based on the theory of information asymmetry,it gives a theoretical explanation for the matching raito of the lending platform.For the traditional risk control model,we focused on the analysis of its objectivity of the designing and the exporation of underlying information,which leads to the solution of the above problems.Regarding the current situaion of the first problem,the dilemma caused by data diversity and information asymmetry,a credit evaluation model of P2 P online lending platform based on big data analysis is proposed in this studey.Based on traditional credit indexes,the quality and the caliber of users are added to the the assessment to get a more comprehensive evaluation model.The method of data mining has been used to integrate more comprehensive user information and valuable indexes can be screened through the information divergence.Based on this model,an example has been performed to verify the feasibility and effectiveness of the risk evaluation model in this study.As for the second problem,that is,the objectivity and the underlying correlation of the traditional evaluation model,a risk control analysis model on the fundation of the research of big data is proposed based on deep learning technology.Using the time sequence to organize data,a deep learning credit evaluation model has been constructed on the basis of the convolutional neural network algorithm.In order to verify the performance of this model,the training-optimized model is compared with logistic regression model,decision tree model and random forest model,using ROC curves and AUC values as evaluation indicators.The results show that the deep learning model possesses higher accuracy,which means that the deep learning model has obvious advantages when analyzing complex nonlinearity subjects with multiple feature.更多还原