分类器集成差异性研究

Survey of diversity researches on classifier ensembles

差异性是集成学习中的重要概念,对差异性的研究在集成学习领域中占有基础性地位。从差异度量方法、差异度与分类器集成性能的关系以及差异度在集成优化中的应用三个方面对当前研究进展进行分析。深入分析了现有工作,对存在的问题给出一些解决思路,建议不能为了差异性以较大的基分类器精度损失为代价;不能为了引入差异性而偏离原来的分类问题。

Diversity is an important concept in ensemble learning and researches on diversity are the foundation of ensemble learning.To the classification problems,progress of ensemble diversity are investigated on three aspects,i.e.diversity measurements,relationship between ensemble diversity and ensemble performance and the application of diversity in optimization of classifier ensembles.Based on going deep into existing works,some resolutions for problems in existence are presented.