Asymptotically Bias-Corrected Regularized Linear Discriminant Analysis for Cost-Sensitive Binary Classification

Amin Zollanvari, Muratkhan Abdirash, Aresh Dadlani, Berdakh Abibullaev

Research output: Contribution to journalArticle

Abstract

In this letter, the theory of random matrices of increasing dimension is used to construct a form of regularized linear discriminant analysis (RLDA) that asymptotically yields the lowest overall risk with respect to the bias of the discriminant in cost-sensitive classification of two multivariate Gaussian distributions. Numerical experiments using both synthetic and real data show that even in finite-sample settings, the proposed classifier can uniformly outperform RLDA in terms of achieving a lower risk as a function of regularization parameter and misclassification costs.

Original languageEnglish
Article number8720003
Pages (from-to)1300-1304
Number of pages5
JournalIEEE Signal Processing Letters
Volume26
Issue number9
DOIs
Publication statusPublished - Sep 1 2019

    Fingerprint

Keywords

  • bias correction
  • cost-sensitive classification
  • random matrix theory
  • Regularized linear discriminant

ASJC Scopus subject areas

  • Signal Processing
  • Electrical and Electronic Engineering
  • Applied Mathematics

Cite this