计算机科学与探索 ›› 2017, Vol. 11 ›› Issue (12): 1993-2003.DOI: 10.3778/j.issn.1673-9418.1610058

• 人工智能与模式识别 • 上一篇    下一篇

基于双模糊信息的特征选择算法

李素姝+,王士同,李  滔   

  1. 江南大学 数字媒体学院,江苏 无锡 214122
  • 出版日期:2017-12-01 发布日期:2017-12-07

Feature Selection Algorithm Based on Doubly Fuzziness Information

LI Sushu+, WANG Shitong, LI Tao   

  1. School of Digital Media, Jiangnan University, Wuxi, Jiangsu 214122, China
  • Online:2017-12-01 Published:2017-12-07

摘要: 在对传统特征选择算法进行研究的基础上,提出了一种基于双模糊信息的特征选择算法(feature    selection algorithm based on doubly fuzziness information,FSA-DFI)。第一种模糊体现在对最小学习机(least learning machine,LLM)进行模糊化后得到模糊最小学习机(fuzzy least learning machine,FUZZY-LLM)中;另一种模糊则是在基于贡献率模糊补充这一方法中体现的,其中贡献率高的特征才可能被选入最终的特征子集。算法FSA-DFI是将FUZZY-LLM和基于贡献率的模糊补充方法结合得到的。实验表明,和其他算法相比,所提特征选择算法FSA-DFI能得到更好的分类准确率、更好的降维效果以及更快的学习速度。

关键词: 特征选择, 双模糊, 最小学习机, 模糊隶属度, 模糊补充

Abstract: Based on the traditional feature selection algorithms, this paper proposes a new feature selection algorithm based on doubly fuzziness information (FSA-DFI). One fuzziness is involved in the fuzzified version of least learning machine (LLM), i.e., FUZZY-LLM, and the other fuzziness is involved in the contribution-rate based fuzzy complementary method in which the features with high contribution rates may be finally selected. The proposed feature selection algorithm is built on the combination of FUZZY-LLM and the contribution-rate based fuzzy complementary method. Experiments indicate that the proposed feature selection algorithm has better classification accuracy, better dimension reduction and faster learning speed, in contrast to other comparative algorithms.

Key words: feature selection, doubly fuzziness, least learning machine, fuzzy membership degree, fuzzy complementary