Credit Risk Management Research Of E-business Based On Naive Bayes Model

The Naive Bayes correlation models are used to classify credit risk data sets of E-business.According to results of the classification,this paper puts forward some measures for both parties to deal with credit risk of E-business so as to promote healthy development of E-business.This paper aims to strengthen classification guidance of credit risk of E-business and improve reliability of credit risk analysis.Because of continuity and redundancy of the E-business credit risk data sets,the Gaussian Naive Bayesian model based on Kernel Principal Component Analysis(KPCA-Gaussian NB)is established to classify different E-business credit risk data sets.By analyzing advantages and disadvantages of the KPCA-Gaussian NB,this paper found that the KPCA-Gaussian NB model can not be applied to all data sets.The reason is that the class labels of some data sets are unevenly distributed.Secondly,the ReliefF feature selection algorithm that was widely used in the binary classification problem is introduced.This paper puts forward a new feature selection algorithm by improving the ReliefF algorithm.The new feature selection algorithm sorts features of the E-business credit risk data sets while retaining more sufficient information and select appropriate feature subset.The Tree Augmented Naive Bayesian algorithm is used as search algorithm to improve classification efficiency of the Naive Bayesian model.It relaxes the strong feature independence assumption of the Naive Bayesian model,the network structure with highest Bayesian score is searched as final classification model.Finally,feasibility and validity of the models are tested by classifying the E-business credit risk data sets.On two levels of evaluation about the summary and the detailed accuracy by class,different indicators are selected as evaluation criterion to evaluate the models.The evaluation results of the models show that the improved ReliefF feature selection algorithm combined with the Tree Augmented Naive Bayes algorithm is better than the original model and other combined models on the whole.