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

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
Pages (from-to)1300-1304
JournalIEEE Signal Processing Letters
Volume26
Issue number9
Publication statusPublished - 2019

Fingerprint

Binary Classification
Discriminant analysis
Discriminant Analysis
Misclassification
Gaussian distribution
Multivariate Distribution
Regularization Parameter
Costs
Random Matrices
Discriminant
Lowest
Classifiers
Classifier
Numerical Experiment
Experiments
Form

Keywords

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

ASJC Scopus subject areas

  • Artificial Intelligence
  • Signal Processing

Cite this

@article{6e6cb70d74254d6b887ad7e39a1e086d,
title = "Asymptotically Bias-Corrected Regularized Linear Discriminant Analysis for Cost-Sensitive Binary Classification",
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.",
keywords = "Regularized linear discriminant, random matrix theory, cost-sensitive classification, bias correction",
author = "Amin Zollanvari and Muratkhan Abdirash and Aresh Dadlani and Berdakh Abibullaev",
year = "2019",
language = "English",
volume = "26",
pages = "1300--1304",
journal = "IEEE Signal Processing Letters",
issn = "1070-9908",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
number = "9",

}

TY - JOUR

T1 - Asymptotically Bias-Corrected Regularized Linear Discriminant Analysis for Cost-Sensitive Binary Classification

AU - Zollanvari, Amin

AU - Abdirash, Muratkhan

AU - Dadlani, Aresh

AU - Abibullaev, Berdakh

PY - 2019

Y1 - 2019

N2 - 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.

AB - 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.

KW - Regularized linear discriminant

KW - random matrix theory

KW - cost-sensitive classification

KW - bias correction

M3 - Article

VL - 26

SP - 1300

EP - 1304

JO - IEEE Signal Processing Letters

JF - IEEE Signal Processing Letters

SN - 1070-9908

IS - 9

ER -