Volume no :5, Issue no: 1, September (2016)

LEAST SQUARES UNIVERSUM TSVM

Author's: Yanmeng Li and Liya Fan
Pages: [1] - [13]
Received Date: November 22, 2015
Submitted by: Mehmet Koc.
DOI: http://dx.doi.org/10.18642/ijamml_7100121598

Abstract

Supervised learning problem with Universum data is a new research subject in machine learning. Universum data, which are not belonging to any class of the classification problem of interest, has been proved very helpful in learning. For data classification with Universum data, a novel quick classifier is proposed in this paper and named as least squares Universum twin support vector machine (LS-U-TSVM). The main advantage of the proposed method is that the running time is shorten greatly by using least squares technique. Experiment results indicate that the proposed LSU-TSVM is an effective and competitive classifier for data classification with Universum data.

Keywords

twin support vector machine, Universum data, least squares technology, classification accuracy.