基于C4.5和NB混合模型的数据流分类算法  

Classification Algorithm for Data Stream Based on Mixture Models of C4.5 and NB

作　　者：李燕[1] 张玉红[1] 胡学钢[1] 

机构地区：[1]合肥工业大学计算机与信息学院,合肥230009

出　　处：《计算机科学》2010年第12期138-142,共5页Computer Science

基　　金：国家973重点基础研究发展计划(2009CB326203);国家自然科学基金课题(60975034);安徽省自然科学基金课题(090412044)资助

摘　　要：具有概念漂移的含噪数据流的分类问题成为数据流挖掘领域研究的热点之一。提出了一种基于C4.5和Nave Bayes混合模型的数据流分类算法CDSMM。它以C4.5作为基分类器,采用朴素贝叶斯分类器过滤噪音,同时引入假设检验中的μ检验方法检测概念漂移,动态更新模型。实验结果表明,CDSMM算法在处理带有噪音的概念漂移数据流时具有比同类算法更好的分类正确率。Classification on the noisy data stream with concept drifts has recently become one of the most popular topics in streaming data mining.Classification algorithm for mining Data Streams based on Mixture Models of C4.5 and NB was proposed,called CDSMM,in which decision trees based on C4.5 are selected as the basic classifiers and the classifier of Nave Bayes is adopted to filter noise data.Meanwhile,it introduces the μ-hypothesis testing method to detect concept drifts.Extensive studies demonstrate that CDSMM is superior to several existing algorithms in the predictive accuracy when handling noisy data streams with concept drifts.

关 键 词：数据流 概念漂移 分类 噪音 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]