Research On Personal Credit Assessment And Risk Warning Of Internet Credi

In recent years,Internet finance industry has been booming.From the perspective of business mode,Internet finance can be divided into four types: Internet credit,Internet wealth management,third-party payment,and crowdfunding in China.With the influence of inclusive finance policy,Internet credit has been promoted rapidly.On the one hand,loan consumption is conducive to expanding household consumption and driving economic growth.On the other hand,excessive loan consumption will bring hidden risk and even lead to banking crisis.Therefore,credit risk management plays a very important role in the development of Internet financial industry.According to the process,credit risk management includes two stages: pre loan and post loan.Strengthening both the two stages of credit risk management is an urgent need to improve the quality of credit assets and ensure the steady development of credit business.In the era of big data,the data amount in the Internet finance industry is increasing,and the data update is accelerating.However,due to the large amount of data and high dimension of variables,Internet credit has changed the traditional way of personal credit risk assessment and brought many challenges to personal credit risk assessment.The paper focuses on the personal credit risk accessment in the internet credit and covers the following contents:1.Research on feature engineering method based on the combination of feature constructionWithin the Internet credit context,the amount of data is large and the dimension of variables is high,but lots of them are weak variables.Aiming at the problem of insufficient information obtained from weak variables,this paper proposes a personal credit assessment model based on the combination of feature construction.The knowledge backed feature construction and the symbol conversion backed feature construction are respectively used to explore the weak variables,extract the high-level features,and enhance the influence of the weak variables.At the same time,the Boruta method,which focuses on the related features of dependent variables,is used to extract the features of high-dimensional data,which helps us to better understand the influencing factors of dependent variables,and reduce the computational cost of credit data.The sample data of this study are the real personal credit data from the bank of China,Central bank credit system and third-party credit investigation agency.Meanwhile,eight machine learning methods are used for empirical research.The results show that the method of feature combination proposed can effectively improve the effect of personal credit risk assessment,which is valuable to the application field of personal credit risk assessment.At the same time,it is found that personal loan behavior and its variation trend have the most significant impact on personal credit risk,also loan strength and repayment pressure have impacts to some degree.2.Personal credit risk assessment based on Bayesian hyperparameter optimization algorithm Light GBM method.Internet credit platforms are striving to ensure that loan request can be approved quickly and provide convenient service to enhance market competitiveness.However,the massive and high-dimensional data poses challenges for credit accessment.From the perspective of hyperparameter optimization of machine learning algorithms,this paper proposes a pre-loan personal credit risk assessment model based on Bayesian hyperparameter optimization algorithm Light GBM method.By introducing a new boosting method—Light GBM,which can solve the time-consuming problem of data in large samples.Its advantages include: fast training speed,high accuracy,low memory,parallel computing supporting,and it can process large data sets.Light GBM contains a variety of super parameters.In this paper,Bayesian hyperparameter optimization is adopted for tuning,which can improve the performance and accuracy of credit risk assessment at the same time.This paper conducts empirical research based on the open data set of Lending club,the data set of the bank of China and the data set of combined construction features.In order to evaluate the proposed personal credit risk assessment method,this paper uses eleven kinds of machine learning algorithms.And because the Internet credit data is unbalanced,this paper uses five kinds of processing methods,including original data,random undersampling,random oversampling,SMOTE and mixed sampling,to explore a comprehensive and accurate credit assessment model.From the two aspects of classification effect and training time,three groups of the experimental results show that the classification of the integrated model has obvious advantages in effect and performance compared with the single model,and the proposed method in this paper takes the least time with high classification effect ranks,which proved this method is feasible and practical in the field of Internet credit personal credit assessment.3.Identify and warn the personal credit risk after loan based on survival analysis.At present,most of the research on post-loan credit management is qualitative.According to the research status of post-loan personal credit risk management,this paper firstly analyzes the personal credit assessment index system after loan,and uses the survival analysis method to establish the personal credit risk assessment model,focusing on post-loan risk influence factors,and predict the outcome and time of eventual default after the initial personal loan to help the individual credit risk early warning.And the paper put forward corresponding policy suggestions at last.Based on the credit data from the bank of China,this paper conducts an empirical study.The results show that personal history credit and post-loan borrowing behaviors have relatively important impacts on the default prediction of borrowers.In particular,both of the borrower’s repayment behavior and the behavior time is very important.From the perspective of risk prevention and risk reduction,if the bank can predict the possibility of non-performing loans as early as possible according to certain information presented by the borrower after loan,it can timely take targeted measures to deal with and prevent them,thus reducing the actual incidence of non-performing loans.As a result,it has good theoretical and practical significance.In summary,with the rapid development of the Internet financial industry,credit risk assessment plays a vital role.Based on big data,this paper uses machine learning and statistical methods to study the pre-loan credit risk assessment and post-loan early warning of Internet credit.The research of this paper has good theoretical and practical significance.