The Application Of Ensemble Support Vector Machine Based On Bagging Algorithm In Personal Credit Rating

Credit has a very important position in economic life.With the development of China's economy,the ability of residents to take financial risks has increased and personal consumer credit is also booming.The credit means money times is coming.At the same time,because China has a large population,the credit data of Chinese consumers are also growing in an explosive manner.The premise of applying personal statistical credit rating methods has matured.we can use statistical methods combined with data mining technologies that are now very popular to research it.The traditional quantitative analysis is not only inefficient,but also depends on the subjective judgment of the rater in many cases.The application of the credit scorecard model changes this situation.The characteristics of credit applicants are quantified,and a series of objective criteria are established.However,there are still many difficulties in the credit rating issue.Whether the applicant will default or not will not only depend on the ability to repay the loan,but will also be effected by the willingness.The consumer's willingness is often not quantified,so it has led to a big error rate in resolving credit rating problems.Support Vector Machine(SVM)had been presented in the 1990s,which is a machine learning method based on statistical learning theory.It not only has excellent learning ability to small sample,but also to solves nonlinear,high-dimensional,over-fit and other machine learning has a significant effect.Ensemble learning is based on training multiple sub-classifiers and synthesizing the decisions of sub-classifiers to make the final decision.This method has improved the generalization ability of the classifier and achieved more accurate results than a single classifier.This paper use Taiwan's credit data.We firstly compares it with the earlier discrim-inant analysis method such as Logistics model and support vector machine method.In terms of the overall accuracy rate,the correct rate of the radial basis kernel function support vector machine is the highest and the discriminant analysis method is the low-est,but the error of the discriminant analysis method is mainly to misclassify a good credit sample into a worse credit sample,which proves that the discriminant analysis method still has value in practical application.Then this paper attempts to use 21-fold sampling and random sampling method using the radial basis kernel function support vector machine to learning and the result show that this method made some improve-ments,but our improvement is very limited.so far,credit rating issues are difficult in the classification problem and there is still a long way to research a method to solve it.