计算机科学与探索 ›› 2011, Vol. 5 ›› Issue (1): 59-67.DOI: 10.3778/j.issn.1673-9418.2011.01.006

• 学术研究 • 上一篇    下一篇

最大化边际的分类器选取算法

付 彬, 王志海, 王中锋   

  1. 北京交通大学 计算机与信息技术学院, 北京 100044
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2011-01-01 发布日期:2011-01-01
  • 通讯作者: 付 彬

Algorithm of Classifier Selection for Maximizing the Margin

FU Bin+, WANG Zhihai, WANG Zhongfeng   

  1. School of Computer and Information Technology, Beijing Jiaotong University, Beijing 100044, China
  • Received:1900-01-01 Revised:1900-01-01 Online:2011-01-01 Published:2011-01-01
  • Contact: FU Bin

摘要: 在分析了不同的多样性定义的基础上, 给出了多样性度量应该考虑的三种因素。结合边际概念, 提出了一种新的多样性度量标准。实验结果表明, 与当前已有的典型的多样性定义相比, 在利用爬山法进行分类器选取时, 使用该方法所选出的分类器子集在大多数数据集合上都有更好的分类性能。

关键词: 集成学习, 分类器, 基分类器, 多样性, 分类器组合

Abstract: After analyzing different definitions of diversity, this paper firstly proposes three factors that should be considered when measuring the diversity. Then it presents a new definition and measurement criterion of diversity,taking the concept of margin into consideration. In the end, the experimental results show that when used in hill-climbing algorithm, this new criterion can pick out classifiers subset with higher accuracy, compared with existed classical definitions of diversity and other criterion.

Key words: ensemble learning, classifier, base classifier, diversity, classifiers combination

中图分类号: