Home > Archive > 2016 > Volume 6 Number 1 (Feb. 2016) >
IJMLC 2016 Vol.6(1): 36-41 ISSN: 2010-3700
DOI: 10.18178/ijmlc.2016.6.1.568

Unsupervised Feature Selection with Correlation and Individuality Analysis

Xiucai Ye, Kaiyang Ji, and Tetsuya Sakurai

Abstract—Feature selection is an important technique for data dimension reduction. Embedded method with sparse regression is wildly used for unsupervised feature selection. The embedded method aims to find a better feature subset by exploiting feature correlation without considering the importance of each feature individually. In this paper, we propose a framework for unsupervised feature selection based on the embedded and spare regression model. Our framework not only exploits the correlation of the features but also analyzes the importance of each individual feature. By using the weight of individual feature to optimize the sparse regression in the process of embedding, the correlation and local structure preserving property of the selected features can be well balanced. We evaluate the proposed framework by using four public datasets. The experimental results demonstrate the superior performance of the proposed framework.

Index Terms—Unsupervised feature selection, embedding method, sparse regression, correlativity, individuality, local preserving.

The authors are with the Computer Science, University of Tsukuba, Tsukuba, Japan (e-mail: yexiucai@mma.cs.tsukuba.ac.jp, jikaiyang@mma.cs.tsukuba.ac.jp, sakurai@cs.tsukuba.ac.jp).

[PDF]

Cite: Xiucai Ye, Kaiyang Ji, and Tetsuya Sakurai, "Unsupervised Feature Selection with Correlation and Individuality Analysis," International Journal of Machine Learning and Computing vol.6, no. 1, pp. 36-41, 2016.

General Information

  • ISSN: 2010-3700 (Online)
  • Abbreviated Title: Int. J. Mach. Learn. Comput.
  • Frequency: Bimonthly
  • DOI: 10.18178/IJMLC
  • Editor-in-Chief: Dr. Lin Huang
  • Executive Editor:  Ms. Cherry L. Chen
  • Abstracing/Indexing: Inspec (IET), Google Scholar, Crossref, ProQuest, Electronic Journals Library.
  • E-mail: ijmlc@ejournal.net


Article Metrics