Research On Personal Credit Evaluation Model Optimized By SVM Based On Binary GWO Algorithm

With the change of people’s consumption consciousness,personal credit business is also gradually expanding.While lending institutions obtain income,there are also huge business risks.Therefore,in order to avoid significant economic losses to lending institutions,we must find a more scientific way to carry out personal credit evaluation.Support vector machine(SVM)has attracted much attention in credit evaluation because it is good at dealing with small data sets,nonlinear and high-dimensional problems.However,there are two factors that will have a great impact on the classification accuracy of SVM,namely feature selection and SVM parameter optimization.Through the summary of a large number of literature,it is found that in the past,when using SVM to solve the problem of personal credit evaluation,the combination of the two is not considered to be optimized at the same time.But in fact,the two affect each other in improving the classification accuracy of SVM,and the improper selection of any factor will reduce the classification accuracy and operation efficiency of SVM.In order to solve this problem,this thesis proposes BGWO-SVM algorithm,which uses binary Gray Wolf Optimizer(BGWO)to select features and optimize SVM parameters at the same time,so as to improve the credit evaluation accuracy of SVM and select fewer features.Then BGWO-SVM is used to solve the problem of personal credit evaluation,and its applicability and effectiveness are verified on the German credit dataset.The main work is showed as following:(1)By drawing the relationship between A1,A3,A4,A6,A19,A20 and credit status,this thesis analyzes the impact of different characteristics on credit status.The results show that feature selection should be performed when dealing with high-dimensional problems to avoid the curse of dimensionality,thereby improving the generalization ability and interpretability of the model.Therefore,this thesis uses the BGWO-SVM algorithm for feature selection to eliminate redundant features,thereby improving the operating efficiency of the model.(2)By establishing a preliminary SVM evaluation model on the credit dataset,it is found that the SVM under the radial basis function(RBF)kernel has better generalization ability.Therefore,this thesis chooses RBF as the kernel function.On 5 datasets from UCI,compared with GA-SVM(without feature selection),GWO-SVM(without feature selection),GA-SVM(with feature selection)and SVM,the accuracy of BGWO-SVM algorithm is improved on average 2.17%-7.78%,compared with GA-SVM(with feature selection),the number of selected features is reduced by 3.82%-14.33%.The experiments show that the parameter optimization and feature selection effect of the BGWO-SVM algorithm is better.(3)Use the BGWO-SVM algorithm to solve the personal credit evaluation problem.On the public German credit dataset,the classification accuracy of BGWO-SVM has been improved on average 4.43%,compared with 6 basic algorithms such as K-nearest neighbor,random forest,and SVM,as well as 3 literature algorithms such as GA-SVM,IBSO-SVM,and BP-AdaBoost.In conclusion,BGWO-SVM can effectively improve the classification accuracy of personal credit evaluation model,so as to provide reliable reference and suggestions for relevant personnel.