一种不平衡数据流集成分类模型  

An Ensemble Classifier Framework for Mining Imbalanced Data Streams

作　　者：欧阳震诤[1] 罗建书[1] 胡东敏[2] 吴泉源[2] 

机构地区：[1]国防科技大学理学院,湖南长沙410073 [2]国防科技大学计算机学院,湖南长沙410073

出　　处：《电子学报》2010年第1期184-189,共6页Acta Electronica Sinica

摘　　要：针对不平衡数据流的分类问题,结合基于权重的集成分类器与抽样技术,本文提出了一种处理不平衡数据流集成分类器模型.理论分析与实验验证表明,该集成分类器具有更低的计算复杂度,更能适应存在概念漂移的不平衡数据流挖掘分类,其整体分类性能优于基于权重的集成分类器模型,能明显提升少数类的分类精度.Many real world data streams mining applications involve learning from imbalanced data streams,where such applications expect to have a higher predictive accuracy over the minority class,however most classification model assume relatively balanced data streams,they cannot handle imbalanced distribution.In this paper,we propose a novel ensemble classifier framework（IMDWE） for mining concept-drifting data streams with imbalanced distribution by using weighted ensemble classifier framework sampling technique including over-sampling and under-sampling.Our empirical study shows that the IMDWE is superior and have improves both the efficiency in learning the model and the accuracy in performing classification over the minority class.

关 键 词：分类 集成分类器 不平衡数据流 概念漂移 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]