·100 智能系统学报 第8卷 漂移的定义、概念漂移的速度与幅度、概念漂移数据 data challenges in the 21st century[J].Science,2011, 流分类器的错误率的界、概念的容量、概念漂移检测 331(6018):700-702 等都还没有得到严格的数学描述，已取得的一些理 [11]STREET W N,KIM Y S.A streaming ensemble algorithm 论上的结果也有待验证和深化， (SEA)for large-scale classification [C]//Proceedings of the Seventh International Conference on Knowledge Discov- 6结束语 ery and Data Mining.San Francisco,USA,2001:377 382. 概念漂移数据流分类给机器学习带来了巨大的 [12]ANGLUIN D.Queries and concept leamning[J].Machine 挑战，本论文从数据流与概念漂移、概念漂移数据流 Learning,1988,2(4):319-342. 分类研究的发展与趋势、概念漂移数据流分类的主 [13 DOMINGOS P,HULTEN G.Mining high-speed data 要研究领域、概念漂移数据流分类研究的新动态4 stream[C]//Proceedings of the Sixth ACM SIGKDD In- 个方面分析总结了学者们在概念漂移数据流分类问 ternational Conference on Knowledge Discovery and Data 题上杰出的研究工作，并分析了当前概念漂移数据 Mining.Boston,USA,2000:71-80. 流分类存在的问题.尽管概念漂移数据流分类已经 [14]HULTEN G,SPENCER L,DOMINGOS P.Mining time- 取得了丰硕的研究成果，且已经成为机器学习研究 changing data streams[C]//Proceedings of the Seventh ACM SIGKDD Intemational Conference on Knowledge Dis- 的热点之一，然而其研究远还没有成熟，还存在着众 covery and Data Mining.San Francisco,USA,2001:97- 多挑战，期待着学术界能有更大的突破： 106. 参考文献： [15 WIDMER G,KUBAT M.Effective learning in dynamic environments by explicit context tracking[C]//Proceed- [1]MITCHELL T M.Machine learning M ]New York: ings of the Sixth European Conference on Machine Leam- McGraw-Hill,1997:15-37. ing.Vienna,Austria,1993:69-101. [2]SCHLIMMER J,GRANGER R.Incremental learning from [16]KELLY M,HAND D,ADAMS N.The impact of changing noisy data[J].Machine Learning,1986,1(3):317-354. populations on classifier performance[C]//Proceedings of [3]GROSSBERG S.Nonlinear neural networks:principles, the Fifth Interational Conference on Knowledge Discovery mechanisms and architecture[J].Neural Network,1988, and Data Mining.San Diego,USA,1999:367-371. 1(1):1761. [17]NARASIMHAMURTHY A,KUNCHEVA L I.A frame- [4]KUNCHEVA L I.Classifier ensembles for changing environ- work for generating data to simulate changing environments ments[C]//Proceedings of the Fifth Workshop on Multiple [C]//Proceedings of the 25th IASTED International Classifier Systems.Cagliari,Italy,2004:1-15 Multi-Conference:Artificial Intelligence and Applications. [5]TSYMBAL A.The problem of concept drift:definitions and Innsbruck,Austria,2007:384-389. related work.TCD-CS-2004-15[R].Dublin:Department [18]MINKU LL,WHITE A P,YAO X.The impact of diversi- of Computer Science Trinity College,2004. ty on online ensemble learning in the presence of concept [6]王涛，李舟军，颜跃进，等。数据流挖掘分类技术综述 drift[J].IEEE Transactions on Knowledge and Data Engi- [J].计算机研究与发展，2007,44(11)：1809-1815. neering,2010,22(5):730-742. WANG Tao,LI Zhoujun,YAN Yuejin,et al.A survey of [19]ELWELL R,POLIKAR R.Incremental leaming of concept classification of data streams[J].Journal of Computer Re- drift in nonstationary environments[J].IEEE Transactions search and Development,2007,44(11):1809-1815. on Neural Networks,2011,22(10):1517-1531. [7]ZLIOBAITE I.Learing under concept drift:an overview [20]MASUD M,GAO J,KHAN L,et al.Integrating novel [EB/OL].[2012-01-12 ]http://zliobaite.googlepages. class detection with classification for concept-drifting data com/Zliobaite_CDoverview.pdf. streams[C]//Proceedings of the European Conference on [8]HOENS T R,POLIKAR R,CHAWLA N V.Learning from Machine Learning and Knowledge Discovery in Databases. streaming data with concept drift and imbalance:an over- Bled,Slovenia,2009:79-94. view[J].Progress in Artificial Intelligence,2012,1(1): 21 MORENO-TORRES J G.RAEDER T,ALAIZ-RODRIGU- 89-101. EZ R.A unifying view on dataset shift in classification [9]GAMA J.A survey on learning from data streams:current [J].Pattemn Recognition,2012,45:521-530. and future trends[J].Progress in Artificial Intelligence, [22]KILANDER F,JANSSON C.COBBIT-a control proce- 2012,1(1):45-55, dure for COBWEB in the presence of concept drift[C]/ [10]OVERPECK J T,MEEHL G A,BONY S,et al.Climate Proceedings of the Sixth European Conference on Machine