基于堆叠集成的数据流分类

Classifying data streams by stacking ensemble

对数据流分类分析的常用方法是集成学习。为了得到更好的分类效果,给出一种基于堆叠集成的数据流分类分析方法。该方法通过构造一个分类器对基分类器进行集成。实验结果表明,与基于投票或加权投票的集成方法相比,基于堆叠集成方法对概念漂移的快速适应能力以及预测准确率得到了提高。

Ensemble learning is a general method for classifying data streams. In order to get a better classification, this paper proposed a general framework for classifying data streams by stacking ensemble. Built another classifier to combine base classifiers. Experiments show that comparing majority vote or weight vote ensemble classifiers, stacking ensemble classifiers has stronger ability in adapting to concept drifting and higher accuracy.