An empirical study of reducing multiclass classification methodologies

R. Kyle Eichelberger; Victor S. Sheng

doi:10.1007/978-3-642-39712-7_39

An empirical study of reducing multiclass classification methodologies

R. Kyle Eichelberger, Victor S. Sheng

Computer Science

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

1 Scopus citations

Abstract

One-against-all and one-against-one are two popular methodologies for reducing multiclass classification problems into a set of binary classifications. In this paper, we are interested in the performance of both one-against-all and one-against-one for basic classification algorithms, such as decision tree, naïve bayes, support vector machine, and logistic regression. Since both one-against-all and one-against-one work like creating a classification committee, they are expected to improve the performance of classification algorithms. However, our experimental results surprisingly show that one-against-all worsens the performance of the algorithms on most datasets. One-against-one helps, but performs worse than the same iterations of bagging these algorithms. Thus, we conclude that both one-against-all and one-against-one should not be used for the algorithms that can perform multiclass classifications directly. Bagging is an better approach for improving their performance.

Original language	English
Title of host publication	Machine Learning and Data Mining in Pattern Recognition - 9th International Conference, MLDM 2013, Proceedings
Pages	505-519
Number of pages	15
DOIs	https://doi.org/10.1007/978-3-642-39712-7_39
State	Published - 2013
Event	9th International Conference on International Conference on Machine Learning and Data Mining, MLDM 2013 - New York, NY, United States Duration: Jul 19 2013 → Jul 25 2013

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	7988 LNAI
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	9th International Conference on International Conference on Machine Learning and Data Mining, MLDM 2013
Country	United States
City	New York, NY
Period	07/19/13 → 07/25/13

Keywords

All-at-Once
C4.5
Logistic Regression
Naive Bayes
One-Against-All
One-Against-One
SVM
multiclass classification

Access to Document

10.1007/978-3-642-39712-7_39

Link to publication in Scopus

Fingerprint Dive into the research topics of 'An empirical study of reducing multiclass classification methodologies'. Together they form a unique fingerprint.

Cite this

Eichelberger, R. K., & Sheng, V. S. (2013). An empirical study of reducing multiclass classification methodologies. In Machine Learning and Data Mining in Pattern Recognition - 9th International Conference, MLDM 2013, Proceedings (pp. 505-519). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 7988 LNAI). https://doi.org/10.1007/978-3-642-39712-7_39

Eichelberger, R. Kyle ; Sheng, Victor S. / An empirical study of reducing multiclass classification methodologies. Machine Learning and Data Mining in Pattern Recognition - 9th International Conference, MLDM 2013, Proceedings. 2013. pp. 505-519 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{6bcacb47e959480db5edc33702f3cae4,

title = "An empirical study of reducing multiclass classification methodologies",

abstract = "One-against-all and one-against-one are two popular methodologies for reducing multiclass classification problems into a set of binary classifications. In this paper, we are interested in the performance of both one-against-all and one-against-one for basic classification algorithms, such as decision tree, na{\"i}ve bayes, support vector machine, and logistic regression. Since both one-against-all and one-against-one work like creating a classification committee, they are expected to improve the performance of classification algorithms. However, our experimental results surprisingly show that one-against-all worsens the performance of the algorithms on most datasets. One-against-one helps, but performs worse than the same iterations of bagging these algorithms. Thus, we conclude that both one-against-all and one-against-one should not be used for the algorithms that can perform multiclass classifications directly. Bagging is an better approach for improving their performance.",

keywords = "All-at-Once, C4.5, Logistic Regression, Naive Bayes, One-Against-All, One-Against-One, SVM, multiclass classification",

author = "Eichelberger, {R. Kyle} and Sheng, {Victor S.}",

year = "2013",

doi = "10.1007/978-3-642-39712-7_39",

language = "English",

isbn = "9783642397110",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

pages = "505--519",

booktitle = "Machine Learning and Data Mining in Pattern Recognition - 9th International Conference, MLDM 2013, Proceedings",

}

Eichelberger, RK & Sheng, VS 2013, An empirical study of reducing multiclass classification methodologies. in Machine Learning and Data Mining in Pattern Recognition - 9th International Conference, MLDM 2013, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 7988 LNAI, pp. 505-519, 9th International Conference on International Conference on Machine Learning and Data Mining, MLDM 2013, New York, NY, United States, 07/19/13. https://doi.org/10.1007/978-3-642-39712-7_39

An empirical study of reducing multiclass classification methodologies. / Eichelberger, R. Kyle; Sheng, Victor S.

Machine Learning and Data Mining in Pattern Recognition - 9th International Conference, MLDM 2013, Proceedings. 2013. p. 505-519 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 7988 LNAI).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - An empirical study of reducing multiclass classification methodologies

AU - Eichelberger, R. Kyle

AU - Sheng, Victor S.

PY - 2013

Y1 - 2013

N2 - One-against-all and one-against-one are two popular methodologies for reducing multiclass classification problems into a set of binary classifications. In this paper, we are interested in the performance of both one-against-all and one-against-one for basic classification algorithms, such as decision tree, naïve bayes, support vector machine, and logistic regression. Since both one-against-all and one-against-one work like creating a classification committee, they are expected to improve the performance of classification algorithms. However, our experimental results surprisingly show that one-against-all worsens the performance of the algorithms on most datasets. One-against-one helps, but performs worse than the same iterations of bagging these algorithms. Thus, we conclude that both one-against-all and one-against-one should not be used for the algorithms that can perform multiclass classifications directly. Bagging is an better approach for improving their performance.

AB - One-against-all and one-against-one are two popular methodologies for reducing multiclass classification problems into a set of binary classifications. In this paper, we are interested in the performance of both one-against-all and one-against-one for basic classification algorithms, such as decision tree, naïve bayes, support vector machine, and logistic regression. Since both one-against-all and one-against-one work like creating a classification committee, they are expected to improve the performance of classification algorithms. However, our experimental results surprisingly show that one-against-all worsens the performance of the algorithms on most datasets. One-against-one helps, but performs worse than the same iterations of bagging these algorithms. Thus, we conclude that both one-against-all and one-against-one should not be used for the algorithms that can perform multiclass classifications directly. Bagging is an better approach for improving their performance.

KW - All-at-Once

KW - C4.5

KW - Logistic Regression

KW - Naive Bayes

KW - One-Against-All

KW - One-Against-One

KW - SVM

KW - multiclass classification

UR - http://www.scopus.com/inward/record.url?scp=84881236873&partnerID=8YFLogxK

U2 - 10.1007/978-3-642-39712-7_39

DO - 10.1007/978-3-642-39712-7_39

M3 - Conference contribution

AN - SCOPUS:84881236873

SN - 9783642397110

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 505

EP - 519

BT - Machine Learning and Data Mining in Pattern Recognition - 9th International Conference, MLDM 2013, Proceedings

Y2 - 19 July 2013 through 25 July 2013

ER -

Eichelberger RK, Sheng VS. An empirical study of reducing multiclass classification methodologies. In Machine Learning and Data Mining in Pattern Recognition - 9th International Conference, MLDM 2013, Proceedings. 2013. p. 505-519. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-642-39712-7_39