Naslov (srp)

Izbor atributa integracijom znanja o domenu primenom metoda odlučivanja kod prediktivnog modelovanja vremenskih serija nadgledanim mašinskim učenjem

Autor

Marković, Ivana P. 1979-

Doprinosi

Stanković, Milena
Stoimenov, Leonid
Stojković, Suzana
Stanković, Miomir
Milovanović, Slavoljub

Opis (eng)

The aim of the research presented within this doctoral dissertation is to develop a feature selection methodology through integrating domain-specific knowledge by applying mathematical methods of decision-making, to improve the feature selection process and the precision of supervised machine learning methods for predictive modeling of time series. To integrate domain-specific knowledge, a multi-criteria decision making method is used, i.e. an analytical hierarchical process proven to be successful in numerous studies carried out to date. This approach was selected because it allows the selection of a set of factors based on their relevance, even in the case of mutually opposite criteria. In predicting the movement of time series, the possibility of integrating feature relevance into support vector machines to improve their prediction accuracy was studied. The proposed methodology was applied as a feature-selection method for the predictive modelling of movement of financial time series. Unlike existing approaches, where the feature selection method is based on a quantitative analysis of the input values, the proposed methodology carries out a qualitative evaluation of the attributes in relation to the prediction domain and represents a means of integrating a priori knowledge of the prediction domain.

Opis (srp)

Biografija autora: list [124];Bibliografija: listovi 108-123 Datum odbrane: 03.05.2018. Data mining – Machine learning

Jezik

srpski

Datum

2017

Licenca

Creative Commons licenca
Ovo delo je licencirano pod uslovima licence
Creative Commons CC BY-NC-ND 2.0 AT - Creative Commons Autorstvo - Nekomercijalno - Bez prerada 2.0 Austria License.

http://creativecommons.org/licenses/by-nc-nd/2.0/at/legalcode