摘要:流数据分类旨在从连续不断到达的流式数据中增量学习一个从输入变量到类标变量的映射函数，以便对随时到达的测试数据进行准确分类.在线学习范式作为一种增量式的机器学习技术，是流数据分类的有效工具.主要从在线学习的角度对流数据分类算法的研究现状进行综述.具体地，首先介绍在线学习的基本框架和性能评估方法，然后着重介绍在线学习算法在一般流数据上的工作现状，在高维流数据上解决"维度诅咒"问题的工作现状，以及在演化流数据上处理"概念漂移"问题的工作现状，最后讨论高维和演化流数据分类未来仍然存在的挑战和亟待研究的方向.



Abstract:The objective of streaming data classification is to learn incrementally a decision function that maps input variables to a label variable, from continuously arriving streaming data, so as to accurately classify the test data that may arrive anytime. The online learning paradigm, as an incremental machine learning technology, is an effective tool for classification of streaming data. This paper mainly summarizes, from the perspective of online learning, the recent development of algorithms for streaming data classification. Specifically, the basic framework and the performance evaluation methodology of online learning are first introduced. Then, the latest development of online learning algorithms for general streaming data, for alleviating the "curse of dimensionality" problem in high-dimensional streaming data, and for resolving the "concept drifting" problem in evolving streaming data are reviewed respectively. Finally, future challenges and promising research directions for classification of high-dimensional and evolving streaming data are also discussed.



PDF全文下载地址:

http://jos.org.cn/jos/article/pdf/5916