Please use this identifier to cite or link to this item: https://hdl.handle.net/10955/570
Full metadata record
DC FieldValueLanguage
dc.contributor.authorCandelieri, Antonio-
dc.contributor.authorGrandinetti, Lucio-
dc.contributor.authorConforti, Domenico-
dc.date.accessioned2014-05-27T09:27:47Z-
dc.date.available2014-05-27T09:27:47Z-
dc.date.issued2014-05-27-
dc.identifier.urihttp://hdl.handle.net/10955/570-
dc.descriptionDottorato di Ricerca in Ricerca Operativa Ciclo XXII, a.a. 2009en_US
dc.description.abstractThis work deals with the development and implementation of a high-level classification framework which combines parameters optimization of a single classifier with classifiers ensemble optimization, through meta-heuristics. Support Vector Machines (SVM) is used for learning while the meta-heuristics adopted and compared are Genetic-Algorithms (GA), Tabu-Search (TS) and Ant Colony Optimization (ACO). Single SVM optimization usually concerns two approaches: searching for optimal parameter values of a SVM with a fixed kernel (Model Selection) or with a linear combination of basic kernels (Multiple Kernel Learning), both approaches have been taken into account. Adopting meta-heuristics avoids to perform time consuming grid-approach for testing several classifier configurations. In particular, starting from canonical formulation of GA, this study proposes some changes in order to take into account specificities of classification learning. Proposed solution has been extensively tested on 8 classification datasets (5 of them are of public domain) providing reliable solutions and showing to be effective. In details, unifying Model Selection, Multiple Kernel Learning and Ensemble Learning on a single framework proved to be a comprehensive and reliable approach, and showing that best solutions have been identified by one of the strategies depending on decision problem and/or available data. Under this respect, the proposed framework may represent a new effective and efficient high-level SVM classification learning strategy.en_US
dc.description.sponsorshipUniversità della Calabriaen_US
dc.language.isoenen_US
dc.relation.ispartofseriesMAT/09;-
dc.subjectRicerca operativaen_US
dc.subjectReti <Modelli>en_US
dc.subjectAlgoritmien_US
dc.titleUn framework di soluzione ad alto livello per problemi di classificazione basato su approcci metaeuristicien_US
dc.typeThesisen_US
Appears in Collections:Dipartimento di Ingegneria Informatica, Modellistica, Elettronica e Sistemistica - Tesi di Dottorato

Files in This Item:
File Description SizeFormat 
Thesis.pdf1,6 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.