摘要： 提出一种基于动态加权的粗糙子空间集成方法EROS-DW。利用粗糙集属性约简方法获得多个特征约简子集，并据此训练基分类器。在分类阶段，根据给定待测样本的具体特征动态地为每个基分类器指派相应的权重，采用加权投票组合规则集成各分类器的输出结果。利用UCI标准数据集对该方法的性能进行测试。实验结果表明，相较于经典的集成方法，EROS-DW方法可以获得更高的分类准确率。

关键词: 集成学习, 粗糙集, 属性约简, 动态加权

Abstract: An approach for dynamic weighting ensemble rough subspace called EROS-DW is proposed. The method first gets diverse attribute reducts based on rough set theory and trains base classifiers according to these reducts correspondingly. Next, it dynamically assigns an appropriate weight to each base classifier based on the characteristic of the input sample. Finally, the outputs of base classifiers are combined through weighted majority voting rule. The performance of EROS-DW is tested on UCI benchmark data sets. Experimental results show that EROS-DW can obtain higher classification accuracies compared with classical ensemble learning methods.

Key words: ensemble learning, rough set, attribute reduction, dynamic weighting

中图分类号: 

• TP391