Título: A INCORPORAÇÃO DO CONHECIMENTO PRÉVIO NA ESTRATÉGIA DE DECISÃO DO APRENDIZADO MULTI-OBJETIVO
Autores: Mendeiros, Talles H.; Braga, Antônio P.; Takahashi, Ricardo H. C.
Resumo: This article presents the results of applying a new approach to decision making on the multiobjective classification. The objective of the strategy-making is to select, within the Pareto-optimal solutions, the learning model to minimize the structural risk (the generalization error). In this new approach has been incorporated a prior knowledge that, when available, allows the choice of a most efficient model. The prior knowledge used in decision making reduces the set of solutions in the space of objectives indicating a region of the Pareto composed of appropriate models to solve the problem. The new strategy was assessed in a classification problem with synthetic data, allowing us to conduct an analysis of efficiency in the decision-making in multiobjective learning.
Palavras-chave: Machine learning; multi-objective optimization; decision making; classification
Código DOI: 10.21528/CBRN2009-075
Artigo em PDF: 075_CBRN2009.pdf
Arquivo BibTex: 075_CBRN2009.bib