Support vector machines for knowledge discovery

Shinsuke Sugaya, Einoshin Suzuki, Shusaku Tsumoto

研究成果: 書籍/レポート タイプへの寄稿会議への寄与

2 被引用数 (Scopus)

抄録

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.

本文言語英語
ホスト出版物のタイトルPrinciples of Data Mining and Knowledge Discovery - 3d European Conference, PKDD 1999, Proceedings
編集者Jan M. Żytkow, Jan Rauch
出版社Springer Verlag
ページ561-567
ページ数7
ISBN(印刷版)3540664904, 9783540664901
DOI
出版ステータス出版済み - 1999
外部発表はい
イベント3rd European Conference on Principles and Practice of Knowledge Discovery in Databases, PKDD 1999 - Prague, チェコ共和国
継続期間: 9月 15 19999月 18 1999

出版物シリーズ

名前Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
1704
ISSN(印刷版)0302-9743
ISSN(電子版)1611-3349

その他

その他3rd European Conference on Principles and Practice of Knowledge Discovery in Databases, PKDD 1999
国/地域チェコ共和国
CityPrague
Period9/15/999/18/99

!!!All Science Journal Classification (ASJC) codes

  • 理論的コンピュータサイエンス
  • コンピュータ サイエンス(全般)

フィンガープリント

「Support vector machines for knowledge discovery」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル