Pengfei Zhao, Xiaoqun He, Yanfei Wang

【Abstract】

Taking into account the feature of credit scoring, this paper firstly improves theoriginalBagging algorithm, and then bulid some bagging ensemble models based on some baseclassifiers. At last, this paper combines the bagging ensemble models in some new way toconstruct a doublelayer ensemblemodel, This paper conducted empirical analysis and cometo the following conclusions: First compared to the original bagging algorithm, the secondprediction error rate of the improved algorithm reduces significantly. Second, compared tosinglelayer model, the doublelayer model reduces the second prediction error rate and the lossrate under the premise of the overall prediction accuracy nondecreasing.

【Keywords】

credit scoring, prediction accuracy, bagging, ensemble model