The combination of multiple classifiers using an evidential reasoning approach

Yaxin Bi, Jiwen Guan, David Bell

Research output: Contribution to journalArticle

79 Citations (Scopus)

Abstract

In many domains when we have several competing classifiers available we want tosynthesize them or some of them to get a more accurate classifier by a combinationfunction. In this paper we propose a ‘class-indifferent’ method for combining classifierdecisions represented by evidential structures called triplet and quartet, using Dempster’srule of combination. This method is unique in that it distinguishes important elementsfrom the trivial ones in representing classifier decisions, makes use of more informationthan others in calculating the support for class labels and provides a practical way toapply the theoretically appealing Dempster–Shafer theory of evidence to the problem ofensemble learning. We present a formalism for modelling classifier decisions as tripletmass functions and we establish a range of formulae for combining these mass functionsin order to arrive at a consensus decision. In addition we carry out a comparativestudy with the alternatives of simplet and dichotomous structure and also compare twocombination methods, Dempster’s rule and majority voting, over the UCI benchmark data,to demonstrate the advantage our approach offers.
LanguageEnglish
Pages1731 -1751
JournalArtificial Intelligence
Volume172
Issue number15
DOIs
Publication statusPublished - Oct 2008

Fingerprint

Classifiers
voting
Labels
learning
evidence
Evidentials
Classifier

Cite this

Bi, Yaxin ; Guan, Jiwen ; Bell, David. / The combination of multiple classifiers using an evidential reasoning approach. 2008 ; Vol. 172, No. 15. pp. 1731 -1751.
@article{0c9cc9ae909a4d7a9bf536d69b3e303f,
title = "The combination of multiple classifiers using an evidential reasoning approach",
abstract = "In many domains when we have several competing classifiers available we want tosynthesize them or some of them to get a more accurate classifier by a combinationfunction. In this paper we propose a ‘class-indifferent’ method for combining classifierdecisions represented by evidential structures called triplet and quartet, using Dempster’srule of combination. This method is unique in that it distinguishes important elementsfrom the trivial ones in representing classifier decisions, makes use of more informationthan others in calculating the support for class labels and provides a practical way toapply the theoretically appealing Dempster–Shafer theory of evidence to the problem ofensemble learning. We present a formalism for modelling classifier decisions as tripletmass functions and we establish a range of formulae for combining these mass functionsin order to arrive at a consensus decision. In addition we carry out a comparativestudy with the alternatives of simplet and dichotomous structure and also compare twocombination methods, Dempster’s rule and majority voting, over the UCI benchmark data,to demonstrate the advantage our approach offers.",
author = "Yaxin Bi and Jiwen Guan and David Bell",
year = "2008",
month = "10",
doi = "10.1016/j.artint.2008.06.002",
language = "English",
volume = "172",
pages = "1731 --1751",
number = "15",

}

The combination of multiple classifiers using an evidential reasoning approach. / Bi, Yaxin; Guan, Jiwen; Bell, David.

Vol. 172, No. 15, 10.2008, p. 1731 -1751.

Research output: Contribution to journalArticle

TY - JOUR

T1 - The combination of multiple classifiers using an evidential reasoning approach

AU - Bi, Yaxin

AU - Guan, Jiwen

AU - Bell, David

PY - 2008/10

Y1 - 2008/10

N2 - In many domains when we have several competing classifiers available we want tosynthesize them or some of them to get a more accurate classifier by a combinationfunction. In this paper we propose a ‘class-indifferent’ method for combining classifierdecisions represented by evidential structures called triplet and quartet, using Dempster’srule of combination. This method is unique in that it distinguishes important elementsfrom the trivial ones in representing classifier decisions, makes use of more informationthan others in calculating the support for class labels and provides a practical way toapply the theoretically appealing Dempster–Shafer theory of evidence to the problem ofensemble learning. We present a formalism for modelling classifier decisions as tripletmass functions and we establish a range of formulae for combining these mass functionsin order to arrive at a consensus decision. In addition we carry out a comparativestudy with the alternatives of simplet and dichotomous structure and also compare twocombination methods, Dempster’s rule and majority voting, over the UCI benchmark data,to demonstrate the advantage our approach offers.

AB - In many domains when we have several competing classifiers available we want tosynthesize them or some of them to get a more accurate classifier by a combinationfunction. In this paper we propose a ‘class-indifferent’ method for combining classifierdecisions represented by evidential structures called triplet and quartet, using Dempster’srule of combination. This method is unique in that it distinguishes important elementsfrom the trivial ones in representing classifier decisions, makes use of more informationthan others in calculating the support for class labels and provides a practical way toapply the theoretically appealing Dempster–Shafer theory of evidence to the problem ofensemble learning. We present a formalism for modelling classifier decisions as tripletmass functions and we establish a range of formulae for combining these mass functionsin order to arrive at a consensus decision. In addition we carry out a comparativestudy with the alternatives of simplet and dichotomous structure and also compare twocombination methods, Dempster’s rule and majority voting, over the UCI benchmark data,to demonstrate the advantage our approach offers.

U2 - 10.1016/j.artint.2008.06.002

DO - 10.1016/j.artint.2008.06.002

M3 - Article

VL - 172

SP - 1731

EP - 1751

IS - 15

ER -