P2P网络借贷违约风险预测模型实证研究

Empirical Research on Default Risk Prediction Model of P2P Network Lending

摘要: 本文的实证研究数据来自Kaggle网站的比赛数据，该数据集爬取自某上市公司的用户信息，主要包含了6万多个借贷人的情况及贷款状态(是否违约)。本文依次基于传统的logistic回归模型、贝叶斯决策树、支持向量机和随机森林算法构建违约预测模型。按照文章中构建的评估指标来进行比较，得到随机森林模型的预测效果最佳。结果表明，本文选出的影响客户信用好坏的特征和风险预测模型有解释性，可以通过随机森林模型来预测客户的违约风险，有利于P2P网贷的发展同时也极大地为借贷公司减小损失。

Abstract: The empirical research data of this paper is from the contest data of Kaggle website. The data set is extracted from the user information of a listed company, mainly including the situation and loan status (default or not) of more than 60,000 borrowers. This paper constructs default prediction model based on traditional Logistic regression model, Bayesian decision tree, support vector machine and random forest algorithm successively. According to the evaluation indexes constructed in this paper, the prediction effect of random forest model is the best. The results show that the characteristics and risk prediction models selected in this paper are explanatory, and the default risk of customers can be predicted through the random forest model, which is conducive to the develop-ment of P2P lending network and greatly reduces the loss of lending companies.