Selecting the best hyperplane in the framework of optimal pairwise linear classifiers

  • Authors:
  • Luis Rueda

  • Affiliations:
  • School of Computer Science, University of Windsor, 401 Sunset Avenue, Windsor, ON, Canada N9B 3P4

  • Venue:
  • Pattern Recognition Letters
  • Year:
  • 2004

Quantified Score

Hi-index 0.10

Visualization

Abstract

In this paper, we introduce a new approach to selecting the best hyperplane from the pairwise classifier (BHPC) when the optimal pairwise linear classifier is given. We first propose a procedure for selecting the BHPC, and analyze the conditions in which the BHPC is selected. In one of the cases, it is formally shown that the BHPC and Fisher's classifier (FC) are coincident. To evaluate the efficiency of the new classifier, we present an empirical and graphical analysis on synthetic data and real-life datasets from the UCI machine learning repository, which involves the optimal quadratic classifier, the BHPC, the optimal pairwise linear classifier, and FC. A numerical analysis of the classification error for these classifiers is also included. The results obtained demonstrate that the BHPC is more accurate than FC, and achieves nearly optimal classification.