分类器集成中的多样性度量

分类器之间的多样性被认为是分类器集成的一个关键因素.然而,目前多样性没有统一的定义和度量,也没有确定的操作方法.针对这些问题,总结和介绍了现有的一些多样性度量方法,及其在可视化、构造分类器集成方面的多种具体应用方法.最后,讨论了精度与多样性两难问题,并给出了多样性方法的有效性说明,指出关于集成学习和多样性的研究还有很多问题亟待解决.

Measue Diversity in Classifer Ensembles

Classifier diversity is regarded as a key factor in Classifier Ensemble problem.However,there exists no uniform definition,measurement and operation method of diversity.For these issues,this paper summarized and introduced some existing diversity measurements,as well as their application in visualization and classifier ensemble construction.In the end,the dilemma of diversity and accuracy is discussed,and the effectiveness of diversity is represented,pointing out that lots of problems in the research of ensemble learning and diversity are still waiting to be solved.