TY - GEN
T1 - Support vector machines for knowledge discovery
AU - Sugaya, Shinsuke
AU - Suzuki, Einoshin
AU - Tsumoto, Shusaku
N1 - Publisher Copyright:
© Springer-Verlag Berlin Heidelberg 1999.
Copyright:
Copyright 2020 Elsevier B.V., All rights reserved.
PY - 1999
Y1 - 1999
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
U2 - 10.1007/978-3-540-48247-5_74
DO - 10.1007/978-3-540-48247-5_74
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 - Żytkow, Jan M.
A2 - Rauch, Jan
PB - Springer Verlag
T2 - 3rd European Conference on Principles and Practice of Knowledge Discovery in Databases, PKDD 1999
Y2 - 15 September 1999 through 18 September 1999
ER -