• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2005, Vol. 27 ›› Issue (12): 68-71.

• 论文 • 上一篇    下一篇

特征选择方法综述

王娟[1] 慈林林[2] 姚康泽[2]   

  • 出版日期:2005-01-01 发布日期:2010-06-22

  • Online:2005-01-01 Published:2010-06-22

摘要:

本文总结并提出了较为完备的特征提取定义。根据特征子集形成过程将特征选择分为穷举式、启发式和随机式三类;根据特征评价标准将特征选择分为距离测度、信息测度、相关性测度、一致性测度和分类器错误率五类。通过分析特征选择的影响因素,提出了选择特征、选择方法应该遵循的原则。

关键词: 特征选择 模式识别 特征评价

Abstract:

This paper analyzes and summarizes the previous definition of feature selection,and then introduces a self-contained definition. It divides feature se lection into three classes according to the selecting strategy, and categorizes the methods into five styles by the evaluation function. Through analyzi ng the infection factors in the feature selection technology,this paper introduces some principles to pave the way for practitioners who search for suit able features to solve real-world applications.

Key words: (feature selection, recognition, feature evaluation)