社交数据驱动的P2P借贷风险评估模型

发布时间：2018-01-23 19:33

本文关键词： 互联网金融 P2P借贷风险评估 GBDT 社交　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：P2P借贷作为互联网金融的新兴形式,已经成为个人理财的一个重要选择,但是P2P借贷普遍存在违约问题。针对这一问题,现有的风险评估模型往往只利用借贷的标的信息和借贷人本身信息来降低违约风险。然而不同的借款人之间并不是独立的,因此我们引入了社交的概念(Social),并使用增强学习模型中一类较好的分类器GBDT模型(Gradient Boosting Decision Tree),提出了社交数据驱动的P2P借贷风险评估模型(GBDT-SOC)。首先,我们通过对P2P借贷数据的统计,发现社交因素与借贷人违约之间的潜在关系;分别对借贷人基于朋友信息的点对点联系和基于群组信息的群体间联系建立了相应的社交影响力模型。然后我们进行特征选取和特征的重新构建,并在这些特征上使用GBDT模型进行建模,并将我们建立的社交影响力模型与GBDT模型结合,构建本文的GBDT-SOC模型,对模型进行线性回归。最后利用梯度下降法训练参数,得到对借贷的风险评估结果。本文在P2P借贷平台Prosper的数据集上进行结果分析和对比实验。实验结果表明,相较于传统的P2P借贷风险评估模型,本文在Prosper数据集模型显著地提高了风险预测的准确度。
[Abstract]:P2P lending, as a new form of Internet finance, has become an important choice of personal finance, but P2P lending generally exists default problem. The existing risk assessment models often use only the information of the subject matter of the loan and the information of the borrower to reduce the default risk. However, different borrowers are not independent. So we introduced the concept of social society. A better classifier, GBDT model, is used in the enhanced learning model. A social-data-driven P2P lending risk assessment model is proposed. Firstly, we use the statistics of P2P lending data. Discover the potential relationship between social factors and borrowers' default; The corresponding social influence model is established for the point-to-point connection of the borrower based on the friend information and the inter-group connection based on the group information respectively. Then we carry on the feature selection and the feature reconstruction. Based on these features, we use GBDT model to model, and combine our social influence model with GBDT model to construct the GBDT-SOC model of this paper. The linear regression of the model is carried out. Finally, the gradient descent method is used to train the parameters. The result of risk assessment is obtained. The results are analyzed and compared on the data set of P2P lending platform Prosper. The experimental results show that. Compared with the traditional P2P loan risk assessment model, this paper significantly improves the accuracy of risk prediction in the Prosper data set model.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F713.36;F831.2

【相似文献】