Evaluation And Researsh On Customer Credit Based On Bayesian Network

As we know,Credit and loan is the important part of China's commercial banks,as the main way to make profits and an important component of the financial market,Credit and loan promotes the development of commerce in China to some extent through its unique form.With the progress of the times,the credit systems of major banks in our country are constantly reforming and deepening.However,at present,China's backward credit system greatly restricts the further growth of personal consumer credit.As the most important part of the credit system,the customer credit evaluation system has become a focus both of the society and academia.Many domestic and foreign scholars have put forward many scientific and reasonable customer credit evaluation models,including statistical method such as Logistic regression and Discriminant analysis and Nonstatistical method such as Decision tree and Support vector machine.but there is less artical that scholars use Bayesian network method to assess customers' credit.Personal credit has many uncertainties due to the particularity of its problems.And The Bayesian network has a strong ability to solve uncertain problems through reasoning,Therefore,using Bayesian networks is a good solution for us.This paper establishes three models through the open data “German credit card data set” of machine learning academic recognized institutions,namely,Tree Augmented Naive Bayes Network,Markov blanket and Feature selection tree augmented naive bayes network.Compared with traditional classification models,these three models have better prediction accuracy on the test set.In terms of time complexity,prediction accuracy,the first misclassification rate,the number of required variables,Finally we compare and evaluate the evaluation results of the three model and give the application scenarios that are suitable for these three methods respectively.