Uma proposta de medida de relvância de atributos multivalorados para classificação

• Ano de defesa: 2008
• Autor(a) principal: Lôbo, Mariana Tasca Fontenelle
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Programa de Pós-Graduação em Computação; Computação
• Palavras-chave em Português: Mineração multi-relacional; Seleção de atributos; Atributos multi-valorados; Medidas de relevância; Classificação; Computer science; Multi-relationary mining; Election of attributes; Multi-valorados attributes; Measures of relevance; Classification; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://app.uff.br/riuff/handle/1/18259

An important step in the knowledgediscovery in databases (KDD) process is the attribute selection step, which aims at choosing a subset of attributes that can represent the important information within the data. Most of the attribute selection methods can only handle simple attribute types, such as categorical and numerical. In particular, these methods cannot be applied to multi-valued attributes, which are attributes that take multiple values simultaneously for the same instancein the database. In many real databases, however, multi-valued attributes are present - e. g. the types of books owned by a person may be represented by a multi-valued attribute. This dissertation proposes a relevance measure for multi-valued attributes, which aims at measuring their importance for classification. The proposed measure tahes into account the ability that the attribute has in determining the instance class. In order to evaluate the proposed measure, experiments were several datavbases subimitted to multi-relational classifiers. The experiments show that the resulting accurancy values follow, in most cases, the values of the proposed relevance measure. So we can conclude that the proposed measure is a good indicator of the relevance of multi-valued attributes for classification.