Regularization Of Credit Risk Assessment Of Support Vector Machines

With the increase of credit business in China,the financial institutions and the local economy flourish,but the credit risk is also growing,the so-called credit risk is the borrower cannot repay the debt,resulting in the possibility of bank losses.So we need a scientific and effective way to assess personal credit.This paper will evaluate credit risk based on regularized support vector machines.Support vector machine is a new supervised learning method proposed by Vapnik in statistical learning theory.It has the advantages of global optimal,simple structure and easy generalization,but the standard support vector machine has no ability of feature selection.The addition of regularization is to give the model the ability to choose features.The data for this paper is one microfinance data which comes from a large state-owned bank.Firstly,in the specific study,this paper adds the regularization thought to the support vector machine model,and the Lasso-SVM and Elastic Net-SVM models are established respectively.It is found that the introduction of Lasso method has stronger feature extraction ability and more accurate accuracy than Elastic Net model.Secondly,this paper compares the evaluation results under the regularized support vector machine model with the standard support vector machine model under the whole variable.We found that,same as the support vector machine model,the Lasso-SVM model selected 21 variables.The prediction accuracy of this model not only doesn't decrease because of the decrease of the number of variables,but it is more effective than the kernel function Sigmoid,radial basis and polynomial,which shows that Lasso-SVM has effective feature selection and credit evaluation capability.Finally,this paper compares the model with the well-known Lasso-Logistic model with good predictive results.It is found that the Lasso-SVM model is not only 11 variables less than the Lasso-Logistic model,but also 0.2% higher in the accuracy of the prediction results.Therefore,this paper provides a new effective method for credit risk,and the method has the promotion value.