基于复杂网络和机器学习的P2P用户违约预测

互联网的发展不断冲击着各个行业,P2P行业作为2013年开始兴起的互联网金融中的重要组成部分最近一段时间由于信用违约等原因,给许多用户带来了不小的财产损失.对于P2P行业来说,对用户的信用预测及防范违约风险是事关公司利润的核心问题.本文利用用户手机通讯录之间的包含关系构建社交网络,并从复杂网络的视角加以分析.通过将分析结果转化为机器学习的输入特征,我们用支持向量机的方法挖掘其内在的关联,从而利用用户的社会网络结构性质预测其信用情况.我们的模型基于知名互联网金融公司闪银所提供的大规模脱敏数据,得到了很好的预测效果.

Predicting bad P2P loans with machine learning and complex network algorithm

From 2013,the P2P (peer-to-peer) companies render people easy access to small loans.P2P has become an important part and trend of internet finance industry.Although companies benefit from loan interest,bad loans can be fatal for their future.Hence prediction of bad loans can help those companies avoid loss and thrive.Here we analyze mobile phone contacts from clients of Wecash,a prominent internet finance company.We build a directed network capturing relationship of each client.We then apply a model with machine learning to predict probability of a client failing to repay the loan.Interestingly,network structure and client neighborhood can shed some light on client credit.