Abstract: 【Purpose/significance】Personal default credit risk is one of the main risks faced by online lending platforms,which has at⁃tracted extensive attention from financial institutions.Traditional P2P network lending default risk prediction models usually use histor⁃ical data for modeling,and the objects of the model are mainly historical performance and default factors.The resulting factor selection bias will affect the generalization ability and risk prediction ability of the model【. Method/process】This paper introduces facial feature big data analysis method,uses deep learning technology to automatically extract facial feature variables,integrates it as a new dimen⁃sion into the credit risk evaluation system with historical loan data as the core,and constructs a new type of credit risk prediction model.
【Result/conclusion】The thesis conducts experiments and empirical analysis based on the data set provided by an internet finance com⁃pany in Nanjing,and the results show that the model proposed in this paper is better than the traditional default risk prediction model.【Innovation/limitation】The innovation of this research is that,on the one hand,the role of real borrowers' facial features in predicting loan defaults in the context of Internet credit based on big data analysis methods has been excavated,providing a new data dimension for the Internet credit risk prediction model,and on the other This aspect provides new ideas for using deep learning methods to auto⁃matically recognize and extract facial features in large-scale image data sets.