Support vector machines for knowledge discovery

Shinsuke Sugaya, Einoshin Suzuki, Shusaku Tsumoto

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

2 Citations (Scopus)


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 M. Żytkow, Jan Rauch
PublisherSpringer Verlag
Number of pages7
ISBN (Print)3540664904, 9783540664901
Publication statusPublished - 1999
Externally publishedYes
Event3rd European Conference on Principles and Practice of Knowledge Discovery in Databases, PKDD 1999 - Prague, Czech Republic
Duration: Sept 15 1999Sept 18 1999

Publication series

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


Other3rd European Conference on Principles and Practice of Knowledge Discovery in Databases, PKDD 1999
Country/TerritoryCzech Republic

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • General Computer Science


Dive into the research topics of 'Support vector machines for knowledge discovery'. Together they form a unique fingerprint.

Cite this