Reinforcement learning for rule extraction from a labeled dataset

Vogiatzis, D; Stafylopatis, A

dc.contributor.author	Vogiatzis, D	en
dc.contributor.author	Stafylopatis, A	en
dc.date.accessioned	2014-03-01T01:18:17Z
dc.date.available	2014-03-01T01:18:17Z
dc.date.issued	2002	en
dc.identifier.issn	13890417	en
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/14920
dc.subject	Reinforcement learning	en
dc.subject	Rule extraction	en
dc.subject.other	algorithm	en
dc.subject.other	analytical error	en
dc.subject.other	article	en
dc.subject.other	artificial neural network	en
dc.subject.other	calculation	en
dc.subject.other	classification	en
dc.subject.other	controlled study	en
dc.subject.other	data analysis	en
dc.subject.other	diabetes mellitus	en
dc.subject.other	intermethod comparison	en
dc.subject.other	learning	en
dc.subject.other	normal distribution	en
dc.subject.other	principal component analysis	en
dc.subject.other	priority journal	en
dc.subject.other	probability	en
dc.subject.other	reinforcement	en
dc.subject.other	symbolism	en
dc.subject.other	validation process	en
dc.title	Reinforcement learning for rule extraction from a labeled dataset	en
heal.type	journalArticle	en
heal.identifier.primary	10.1016/S1389-0417(01)00060-2	en
heal.identifier.secondary	http://dx.doi.org/10.1016/S1389-0417(01)00060-2	en
heal.publicationDate	2002	en
heal.abstract	The article introduces a method, which is based on reinforcement learning, for extracting rules of the form if-then-else from a labeled data-set. The constituent parts of a rule are the input dimensions of the labeled data-set, each accompanied by an appropriate interval of activation, and a label which stands for class membership. Initially, the input space is partitioned using tiles. The algorithm tries to compose the largest possible orthogonal intervals out of tiles. After the creation of intervals for each dimension the rule receives credit for its classification ability. This credit with the aid of reinforcement will be used to improve its constituent parts. The effectiveness of the proposed method has been tested on five different classification problems: the Iris data set, the Concentric data, the 4 Gaussians, the Pima-Indians set and the Image Segmentation data set. © 2004 Elsevier Science B.V. All rights reserved.	en
heal.journalName	Cognitive Systems Research	en
dc.identifier.doi	10.1016/S1389-0417(01)00060-2	en
dc.identifier.volume	3	en
dc.identifier.issue	2	en
dc.identifier.spage	237	en
dc.identifier.epage	253	en