摘要： 选择性集成分类算法虽能提高集合分类器在整体数据集上的分类性能，但针对某一具体数据进行分类时，其选择出的个体分类器集合并不一定是最优组合。为此，从数据自适应角度出发，提出一种数据流选择性集成的两阶段动态融合方法，利用待分类数据所在特征空间中的位置，动态选择个体分类器集合，并对其进行分类。理论分析和实验结果表明，与GASEN算法相比，该方法的分类准确率更高。

关键词: 数据流, 选择性集成, 分类, 自适应, 特征空间

Abstract: Selective ensemble classifiers can improve classification accuracy rate of data set. But for a specific data classification, the classifiers contained by ensemble can not be the best combination. Proceed from adaptation of data, two-phase selective ensemble of data streams is presented. According to location of data in the eigenspace, individual classifier is selected to classify this data. Theories and empirical analyses indicate this algorithm has more classification accuracy rate, in contrast with GASEN algorithm.

Key words: data stream, selective integration, classification, self-adaption, eigenspace

中图分类号: 

• N945