Font Size: a A A

Research On Credit Risk Of Enterprise Based On Inference Engine And GBDT

Posted on:2020-11-19Degree:MasterType:Thesis
Country:ChinaCandidate:J H KeFull Text:PDF
GTID:2439330599463054Subject:Finance
Abstract/Summary:PDF Full Text Request
Credit sales model has become a normal enterprise sales model,which will affect the operating costs of enterprises,limit the scale of development of enterprises,and will inevitably lead to uncertain business risks for enterprises that adopt credit sales business model without corresponding credit risk management.How to effectively identify high-quality customers and avoid credit risk has always been the most important problem of credit risk management for enterprises.Firstly,after reading and analyzing the related theories of credit risk management in domestic and foreign literatures,the writer finds that the conclusions drawn from the simple theoretical analysis and the single statistical model analysis of credit risk problems are one-sided,and the statistical model research is usually based on some statistical assumptions,which leads to the conclusion being biased from the actual situation.Secondly,based on the big data tech and machine learning theory,referred to the research conclusions from domestic and foreign literatures,this paper studies the target credit customers of a leading enterprise in an industry.By using internal credit customers as the sample,makes an analysis on historical trade record and the basic information of customers,and to predict the credit risk of customers' overdue credit account.In this study,base data integrate with statistical feature,which transform from temporary flow,learning target credit customer manifestation from time and space dimension and build a multi-level ensemble model.We compared the performance between SVM,GBDT,Logistic Regression and other liner models by studying ACC and predict cost,figured out the optimal fitting model by the same sample data set.It is clearly that hybrid inference machine we introduced has very higher accuracy and very cheaper predicting cost in the case of more input features by compare with traditional form.At the same time,the new ensemble model,which updated traditional liner and multi-layer perception in hybrid inference machine,has extraordinary accuracy to liner model on large data sets but close to liner model on small data sets and more effective to multi-layer perception.In view of the actual business scenario of credit sale in enterprises,this paper deeply analyses the credit management process,evaluates the risk exposures at all stages,puts forward a new credit risk management process based on hybrid inference engine,and ensures the collaboration between the business departments and credit management organization in three stages: pre-prevention,middle-controlling and post-monitoring,so as to cover the main stages and key control point of credit risk process,to form the closed-loop credit management of transaction.
Keywords/Search Tags:Account sale, Credit risk management, Boosting model, Inference engine
PDF Full Text Request
Related items