Support vector machines for knowledge discovery

Shinsuke Sugaya, Einoshin Suzuki, Shusaku Tsumoto

Research output: Chapter in Book/Report/Conference proceedingConference contribution

2 Citations (Scopus)

Abstract

In this paper, we apply support vector machine (SVM) to knowledge discovery (KD) and confirm its effectiveness with a benchmark data set. SVM has been successfully applied to problems in various domains. However, its effectiveness as a KD method is unknown. We propose SVM for KD, which deals with a classification problem with a binary class, by rescaling each attribute based on z-scores. SVM for KD can sort attributes with respect to their effectiveness in discriminating classes. Moreover, SVM for KD can discover crucial examples for discrimination. We settled six discovery tasks with the meningoencephalitis data set, which is a benchmark data set in KD. A domain expert ranked the discovery outcomes of SVM for KD from one to five with respect to several criteria. Selected attributes in six tasks are all valid and useful: their average scores are 3.8-4.0. Discovering order of attributes about usefulness represents a challenging problem. However, concerning this problem, our method achieved a score of more than or equal to 4.0 in three tasks. Besides, crucial examples for discrimination and typical examples for each class agree with medical knowledge. These promising results demonstrate the effectiveness of our approach.

Original languageEnglish
Title of host publicationPrinciples of Data Mining and Knowledge Discovery - 3d European Conference, PKDD 1999, Proceedings
EditorsJan Rauch, Jan M. Żytkow
PublisherSpringer Verlag
Pages561-567
Number of pages7
ISBN (Print)3540664904, 9783540664901
Publication statusPublished - Jan 1 1999
Event3rd European Conference on Principles and Practice of Knowledge Discovery in Databases, PKDD 1999 - Prague, Czech Republic
Duration: Sep 15 1999Sep 18 1999

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume1704
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other3rd European Conference on Principles and Practice of Knowledge Discovery in Databases, PKDD 1999
CountryCzech Republic
CityPrague
Period9/15/999/18/99

Fingerprint

Knowledge Discovery
Support vector machines
Data mining
Support Vector Machine
Attribute
Discrimination
Benchmark
Rescaling
Classification Problems
Sort
Valid
Binary
Unknown
Demonstrate
Class

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Sugaya, S., Suzuki, E., & Tsumoto, S. (1999). Support vector machines for knowledge discovery. In J. Rauch, & J. M. Żytkow (Eds.), Principles of Data Mining and Knowledge Discovery - 3d European Conference, PKDD 1999, Proceedings (pp. 561-567). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 1704). Springer Verlag.

Support vector machines for knowledge discovery. / Sugaya, Shinsuke; Suzuki, Einoshin; Tsumoto, Shusaku.

Principles of Data Mining and Knowledge Discovery - 3d European Conference, PKDD 1999, Proceedings. ed. / Jan Rauch; Jan M. Żytkow. Springer Verlag, 1999. p. 561-567 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 1704).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Sugaya, S, Suzuki, E & Tsumoto, S 1999, Support vector machines for knowledge discovery. in J Rauch & JM Żytkow (eds), Principles of Data Mining and Knowledge Discovery - 3d European Conference, PKDD 1999, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 1704, Springer Verlag, pp. 561-567, 3rd European Conference on Principles and Practice of Knowledge Discovery in Databases, PKDD 1999, Prague, Czech Republic, 9/15/99.
Sugaya S, Suzuki E, Tsumoto S. Support vector machines for knowledge discovery. In Rauch J, Żytkow JM, editors, Principles of Data Mining and Knowledge Discovery - 3d European Conference, PKDD 1999, Proceedings. Springer Verlag. 1999. p. 561-567. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
Sugaya, Shinsuke ; Suzuki, Einoshin ; Tsumoto, Shusaku. / Support vector machines for knowledge discovery. Principles of Data Mining and Knowledge Discovery - 3d European Conference, PKDD 1999, Proceedings. editor / Jan Rauch ; Jan M. Żytkow. Springer Verlag, 1999. pp. 561-567 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{1fb57e45f7484902bc34f224fde7899c,
title = "Support vector machines for knowledge discovery",
abstract = "In this paper, we apply support vector machine (SVM) to knowledge discovery (KD) and confirm its effectiveness with a benchmark data set. SVM has been successfully applied to problems in various domains. However, its effectiveness as a KD method is unknown. We propose SVM for KD, which deals with a classification problem with a binary class, by rescaling each attribute based on z-scores. SVM for KD can sort attributes with respect to their effectiveness in discriminating classes. Moreover, SVM for KD can discover crucial examples for discrimination. We settled six discovery tasks with the meningoencephalitis data set, which is a benchmark data set in KD. A domain expert ranked the discovery outcomes of SVM for KD from one to five with respect to several criteria. Selected attributes in six tasks are all valid and useful: their average scores are 3.8-4.0. Discovering order of attributes about usefulness represents a challenging problem. However, concerning this problem, our method achieved a score of more than or equal to 4.0 in three tasks. Besides, crucial examples for discrimination and typical examples for each class agree with medical knowledge. These promising results demonstrate the effectiveness of our approach.",
author = "Shinsuke Sugaya and Einoshin Suzuki and Shusaku Tsumoto",
year = "1999",
month = "1",
day = "1",
language = "English",
isbn = "3540664904",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "561--567",
editor = "Jan Rauch and Żytkow, {Jan M.}",
booktitle = "Principles of Data Mining and Knowledge Discovery - 3d European Conference, PKDD 1999, Proceedings",
address = "Germany",

}

TY - GEN

T1 - Support vector machines for knowledge discovery

AU - Sugaya, Shinsuke

AU - Suzuki, Einoshin

AU - Tsumoto, Shusaku

PY - 1999/1/1

Y1 - 1999/1/1

N2 - In this paper, we apply support vector machine (SVM) to knowledge discovery (KD) and confirm its effectiveness with a benchmark data set. SVM has been successfully applied to problems in various domains. However, its effectiveness as a KD method is unknown. We propose SVM for KD, which deals with a classification problem with a binary class, by rescaling each attribute based on z-scores. SVM for KD can sort attributes with respect to their effectiveness in discriminating classes. Moreover, SVM for KD can discover crucial examples for discrimination. We settled six discovery tasks with the meningoencephalitis data set, which is a benchmark data set in KD. A domain expert ranked the discovery outcomes of SVM for KD from one to five with respect to several criteria. Selected attributes in six tasks are all valid and useful: their average scores are 3.8-4.0. Discovering order of attributes about usefulness represents a challenging problem. However, concerning this problem, our method achieved a score of more than or equal to 4.0 in three tasks. Besides, crucial examples for discrimination and typical examples for each class agree with medical knowledge. These promising results demonstrate the effectiveness of our approach.

AB - In this paper, we apply support vector machine (SVM) to knowledge discovery (KD) and confirm its effectiveness with a benchmark data set. SVM has been successfully applied to problems in various domains. However, its effectiveness as a KD method is unknown. We propose SVM for KD, which deals with a classification problem with a binary class, by rescaling each attribute based on z-scores. SVM for KD can sort attributes with respect to their effectiveness in discriminating classes. Moreover, SVM for KD can discover crucial examples for discrimination. We settled six discovery tasks with the meningoencephalitis data set, which is a benchmark data set in KD. A domain expert ranked the discovery outcomes of SVM for KD from one to five with respect to several criteria. Selected attributes in six tasks are all valid and useful: their average scores are 3.8-4.0. Discovering order of attributes about usefulness represents a challenging problem. However, concerning this problem, our method achieved a score of more than or equal to 4.0 in three tasks. Besides, crucial examples for discrimination and typical examples for each class agree with medical knowledge. These promising results demonstrate the effectiveness of our approach.

UR - http://www.scopus.com/inward/record.url?scp=84956867156&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84956867156&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:84956867156

SN - 3540664904

SN - 9783540664901

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

SP - 561

EP - 567

BT - Principles of Data Mining and Knowledge Discovery - 3d European Conference, PKDD 1999, Proceedings

A2 - Rauch, Jan

A2 - Żytkow, Jan M.

PB - Springer Verlag

ER -