基于实例加权方法的概念漂移问题研究

Study of example-weighted method for tracking concept drift

数据流上的漂移概念发现已成为数据挖掘领域的研究热点之一。针对存在概念漂移的数据流分类问题,提出一种基于实例加权方法的数据流分类算法(EWAMDS),根据基分类器在训练实例上的分类结果调整该实例的权值,以增强漂移实例在新分类器中的影响,同时引入动态的权值修改因子以提高算法的适应性。实验结果表明,动态地调整实例的权值时算法的适应性更强;与weighted-bagging相比,EWAMDS的时间开销显著降低、分类正确率显著提高。

The tracking of drifting concept from data streams has recently become one of hot spots in data mining.In this paper, a Example-Weighted algorithm for mining data streams （EWAMDS） is proposed for data streams classification in the presence of concept drift,in which weight of train example is adjusted according to base classifier s prediction on it,so as to enhance influence of drifting examples in new classifier,and a dynamic weight modifying factor is introduced to improve the adaptability of this algorithm.The results of experiments indicate that modifying weight of example dynamically makes this algorithm more adaptively;and in comparison with weighted-bagging,EWAMDS has a lower time consumption and higher accuracy.