基于AHP的SMOTEBagging改进模型

An Improved Model of SMOTEBagging Based on AHP

数据不平衡是分类模型在实际应用中常常会遇到的问题,比如信用风险预测、病情诊断等,在这些应用中,提高模型对少类样本的预测准确率有着重要的意义,看重模型的TPR(TruePositiveRate,真正率)表现。SMOTEBagging模型在TPR上比传统Bagging模型表现更好,为了进一步提高其TPR,引入AHP方法对基分类器进行选择性集成,构成了一种新模型,称为AHP-BasedBagging。实验结果表明,AHP-BasedBagging模型能在不牺牲整体预测表现的情况下,以更小的集成规模取得更好的TPR表现,具有更强的实用性。

It often comes with imbalanced data problem when using classification models in the real world applications,such as credit risk prediction and medical diagnosis.In these applications,it is important to improve the accuracy over the minority class,so the performance on the TPR(True Positive Rate)is significant.SMOTEBagging has a better TPR than the normal Bagging model.In order to further improve the TPR of SMOTEBagging,the AHP method is used to selectively integrate the base classifiers and get a novel model,named AHP-Based Bagging.The experimental results show that AHP-Based Bagging can get a better TPR with smaller ensemble size,and not to sacrifice the overall performance,which is more practical.