Research On Credit Risk Early Warning Of Internet Finance Enterprises Based On CNN

Internet finance is a new business model that has provided a useful supplement to the financing difficulties of small and medium-sized enterprises in China’s financial system,while expanding the people’s sources of capital.However,with the development of financial technology and fierce competition between industries,the issue of risks faced by internet finance companies has become increasingly prominent.These risks take various forms,including but not limited to credit risk,information disclosure risk,regulatory risk,compliance risk and so on.Among these risks,credit risk has become an important issue of widespread concern.Therefore,how to timely and effectively warn and assess the credit risk of Internet financial enterprises is necessary to maintain the stability of the financial system.This thesis aims to explore the credit risk early warning model of Internet financial enterprises from the perspective of credit risk management,and carries out systematic research and application demonstration analysis.The research in this thesis covers the following two aspects: firstly,constructing a credit risk early warning index system,and secondly,constructing and applying a credit risk early warning model based on convolutional neural network for Internet financial enterprises.Firstly,this thesis constructs a credit risk early warning indicator system that integrates financial and nonfinancial indicators by considering the incorporation of ESG factors,and uses principal component analysis and grey correlation method to select the indicators,and the final constructed indicator system contains 7 financial dimensional principal factors and 9 nonfinancial indicators,which are used as the input set of the CNN early warning model.Secondly,this thesis illustrates the process of network design and boosting training using CNN neural networks.Considering the sample data characteristics,two subconvolutional networks are designed for the training of dynamic data and static data respectively,and the way for the division of credit ratings and the selection of warning thresholds is explained.Finally,through the simulation application,the 5-year data of 81 listed companies are selected as the samples,and the credit risk status of the sample enterprises is obtained through adaptive synthetic sampling and network adaptive learning.The experimental results show that the sample enterprises are converted into credit scores according to the probability of default,and equal frequency bins are conducted to classify a total of nine credit ratings.The warning threshold is 2.8%,which corresponds to a credit score of 68.7.when the default probability is higher than 2.8%and the credit score is lower than 68.7,the model should issue a warning alert.The highest accuracy rate and the best combined result values of accuracy,recall and F1 value validate the excellent application performance of the model.In summary,this thesis establishes a complete index system and proposes an early warning model based on CNN neural network by systematically studying the credit risk early warning model of Internet financial enterprises.These results have important theoretical value and practical application value for early warning and assessing the credit risk of Internet financial enterprises.