Application Research Of ADABOOST Combination Classification Model In Credit Evaluation

The development of consumer credit has a lot of positive significance for the national society, such as expanding domestic demand, accelerating social consumption structure, improving the structure and the efficiency of bank assets. The key to the development of consumer credit constraints is the lacking of scientific and efficient methods of credit assessment. The main reason to explain why many personal credit scoring models can not be popularized is that the stability of the models is not so good while the accuracy is improved, and the model used by the sample of regional restrictions. This article will introduce the idea of using Adaboost combination of classification in this area, and research the applicability of combined model, to try to resolve the issue.Applicability research mainly consists of three aspects: accuracy, stability and application of classification of models. The review undertaken in this paper covers credit evaluation and combination classification, and analyze the viability of combination classification. Domestic credit sample is used to test the accurability and stability of Adaboost combination classification, and foreign sample are used to test application of the model. From the comparison of empirical result, we find Adaboost combination classification model is better than BP neural network, decision tree and Logistic regression classification model in accuracy 97.33% and stability 0.47%. So Adaboost algorithm improves the situation that we can't obtain accuracy and stability at the same time when using single models. In addition, the accuracy of nonlinear classification model is higher than that of linear models; however, the stability of some nonlinear classification models is worse than that of linear models. In application, the result shows that Adaboost combination classification model can make full use of decision fusion and improve classification performance significantly, compared with the low accuracy of sub-classification model. So it can be concluded that Adaboost combination classification model is practical and have a better application. Although it cannot fit the local credit sample data perfectly, it can improve the classification performance compared with single models especially classification models with poor performance.