流数据分类中的概念漂移问题研究

传统的流数据分类算法基于滑动窗口来优化现有分类器或建立多个分类器来跟踪概念的漂移过程,而不能根据概念漂移的强弱程度自适应地进行分类.在结合当前主流的CVFDT和集成分类器算法的基础之上,提出一种新型流数据分类算法:SADT算法.算法动态地判断概念漂移的发生,自动决定是优化还是重建分类器,适用于不同类型的数据的分类.通过分析和实验论证,该算法在处理概念漂移时具有更好的适应性.

ON CONCEPT DRIFT IN STREAM DATA CLASSIFICATION

Traditional stream data classification approaches use fix-sized sliding windows to optimize current classifier or build several classifiers to trace the drifts of the concepts,but they can not conduct classification self-adaptively according to the level of concept drift.A novel SADT algorithm is proposed based on the integration of current mainstream CVFDT and Weighted Classifier Ensembles algorithms.The new algorithm dynamically analyzes the occurrence of the drifts,and automatically decides to modify or ...