Show simple item record

dc.contributor.authorArrieta, Irati
dc.contributor.authorMendiburu Alberro, Alexander
dc.contributor.authorMori Carrascal, Usue ORCID
dc.contributor.authorLozano Alonso, José Antonio
dc.date.accessioned2020-06-30T10:44:57Z
dc.date.available2020-06-30T10:44:57Z
dc.date.issued2019
dc.identifier.citationEkaia 36 : 291-310 (2019)
dc.identifier.issn0214-9001
dc.identifier.urihttp://hdl.handle.net/10810/44726
dc.description.abstractOne of the most prominent problems in the area of time series data mining is called supervised time series clasication. The goal of this problem is to build a model that predicts the classes of new unclassied series as accurately as possible, departing from a database of time series for which the class is known. As an extension of this problem, on some occasions, the data is collected over time, and, in order to avoid costs that incur in collecting new data or negative consequences that may arise when making late predictions, the goal is to make the class predictions as early as possible. In this context, the problem denominated early classication of time series arises, whose objective is to build a classier that is as accurate as possible, but at the same time, makes the class prediction as early as possible. It is logical to think that the more data points are made available, the more information we have about the time series and, so, it is easier to make accurate class predictions. On the contrary, if we want to make early class predictions, we will have less information and it will be more dicult to make accurate class predictions. Therefore, accuracy and earliness are two objectives which are in con ict. In this work, we propose a innovate method for early classication based on multi-objective formulation of the problem. We have compared it to a model proposed in the literature which models the problem as a single objective optimization problem and we have seen that our model provides better results on some benchmark datasets.; Denborazko serieen datu meatzaritza arloko problema ohikoenetako bat da, denborazko serieen gainbegiratutako sailkapena. Problema honen helburua da, klaseetan banatuta dauden serie multzo batetik abiatuz, sailkatu gabeko beste serie batzuen klasea aurresango duen eredu ahalik eta zehatzena eraikitzea. Problema klasiko honen hedapen gisa, kasu batzuetan, denborazko serieak denboran zehar jasotzen dira, eta ohikoa da iragarpenak ahalik eta lasterren egin nahi izatea, datuak jasotzeak dakartzan kostuak aurrezteko asmoarekin edo klaseak berandu iragartzeak ekarri ditzakeen ondorio kaltegarriak ekiditeko. Egoera honetan, denborazko serieen sailkapen goiztiarra izeneko problema agertzen da, zeinaren helburua den ahalik eta sailkatzaile zehatzena eta aldi berean iragarpen azkarrenak egingo dituena eraikitzea. Nahiko intuitiboa da pentsatzea, seriearen puntu gehiago ditugunean eskuragai, hari buruzko informazio gehiago dugula eta beraz, haren klaseari buruzko iragarpen zehatzagoak burutzea errazagoa dela. Alderantziz, seriearen klasea goiz aurresan nahi badugu, informazio gutxiago izango dugu eta beraz, zailagoa izango da klasea ondo aurresatea. Beraz, zehaztasuna eta azkartasuna bi helburu kontrajarriak dira. Lan honetan, denborazko serieen sailkapen goiztiarrari soluzioa emango dion metodo berritzaile bat proposatzen dugu, helburu anitzeko problema gisa planteaturik. Literaturan aurretik proposatutako helburu bakarreko optimizazio problema gisa eraikitako beste eredu batekin konparatu dugu gure eredua eta ikusi dugu gureak emaitza hobeak ematen dituela zenbait oinarrizko datu-basetan.
dc.language.isoeus
dc.publisherServicio Editorial de la Universidad del País Vasco/Euskal Herriko Unibertsitatearen Argitalpen Zerbitzua
dc.rightsinfo:eu-repo/semantics/openAccess
dc.rights.urihttp://creativecommons.org/licenses/by-nc-sa/4.0/
dc.titleDenborazko serieen sailkapen goiztiarra helburu anitzeko optimizazio problema gisa aztertua.
dc.typeinfo:eu-repo/semantics/article
dc.rights.holder© 2019 UPV/EHU Attribution-NonCommercial-ShareAlike 4.0 International
dc.identifier.doi10.1387/ekaia.19696


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record

© 2019 UPV/EHU Attribution-NonCommercial-ShareAlike 4.0 International
Except where otherwise noted, this item's license is described as © 2019 UPV/EHU Attribution-NonCommercial-ShareAlike 4.0 International