The Research Of Credit Scoring Model For Individual Credit Business

In recent years,Internet finance develops rapidly in China.Many financial institutions are exposed to credit risks.The credit scoring can make use of customers' information to identify potential risks,which plays an essential role in financial institutions.This paper is aimed at the personal credit business.The characteristics high-dimensional of the credit customer data are taken into account by us first.After feature selection,A pre-loan risk-assessment model based on static attributes and FSICA method to achieve the pre-loan risk assessment of credit customers is established.Second,the dynamic attributes of credit customers is used.After time series characterization of dynamic flow data,a middle-loan risk-warning model based on dynamic attributes and ARMA method to achieve the middle-loan risk assessment of credit customers is established.Finally,the two models are combined in parallel to achieve the credit scoring of customers.The contributions of this thesis is described as follows:(1)A feature selection method based on multiple model combined with the search strategy,short for FSICA is proposed.The method combines the advantages of filter,embedded and wrapper.First,two filters and two embedded method are used to comprehensively measure features from feature category correlation and classification ability.Second,Sequential Backward Selection strategy to select feature is used,and the classification accuracy is used as the evaluation index to measure the feature subset.Finally,the optimal feature subset is obtained.The experimental results demonstrate that this method can not only reduce the dimension of feature space,but also improve the classification performance of classification algorithm.(2)A new method for credit scoring using dynamic flow data is proposed.First,the original bank data to get the dynamic flow data is processed.Second,ARMA model based on dynamic flow data and time series theory to obtain the account status of the borrower during the repayment period is established.Third,the balance of the account by the account status of the borrower during the repayment period is done.Finally,the label of the borrower is given by the relationship between the account balance and the loan amount.1694 samples from a bank on this model are used,and the overall accuracy was 90.7? to make sure that our method works effective.The results show that this method has strong feasibility and good application prospects.(3)Combining the above two methods,two risk assessment models: the pre-loan risk-assessment model based on static attributes and FSICA method and the middle-loan risk-warning model based on dynamic attributes and ARMA method are proposed.Finally,the two models are combined to establish a combined model based on pre-loan risk-assessment and middle-loan risk-warning to achieve the risk-assessment of credit customers.The experimental results show that the combined model can be used to evaluate the risk of credit customers before and during loans,and has a good application value.In conclusion,this paper proposes three credit scoring models for personal credit business:the pre-loan risk-assessment model based on static attributes and FSICA method,the middle-loan risk-warning model based on dynamic attributes and ARMA method,the combined model based on pre-loan risk-assessment and middle-loan risk-warning.They are better in assessing the risk of credit customers.It can not only help financial institutions identify customers' qualifications before loan,but also help them find risks reasonably and reduce losses in the process of loan.