Seleção de atributos usando abordagem Wrapper para classificação hierárquica multirrótulo

• Ano de defesa: 2018
• Autor(a) principal: Almeida, Thissiany Beatriz
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Ponta Grossa; Brasil; Programa de Pós-Graduação em Ciência da Computação
• Palavras-chave em Português: Mineração de dados (Computação); Banco de dados; Algorítmos genéticos; Data mining; Data bases; Genetic algorithms; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO; Ciência da Computação
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/3334

The technological advance brings as a consequence a great challenge, which is to find a way to store and extract useful knowledge of a mass of data. An alternative to this scenario is the use of Data Mining tools, being one of the techniques used, the Attribute Selection. The selection of attributes is one of the techniques that can be used to reduce the dimensionality of a database, with the main objective to identify the relevant attributes, thus increasing the predictive power of the classifier. Basically, there are three approaches to Attribute Selection: Embedded, Filter, and Wrapper. In the context of multi-label hierarchical classification, where classes to be predicted are structured according to a hierarchy, few works in the literature present proposals for attribute selection techniques. Thus, in this work, we propose a new method of attribute selection based on the Wrapper approach for global multi-label hierarchical classification. The FSW-HMC method is divided into two main steps: the search strategy (Artificial Immune System and Genetic Algorithm) and subset evaluation (Clus-HMC multi-label hierarchical classifier and AUPRC performance measure). In order to perform the experiments, biological data from 10 Genetic Ontology databases are used, and the classes are structured in a hierarchy in the form of a Directed Acyclic Graph (DAG). When performing the analysis of the generated subsets it can be noted that there was a significant reduction in the number of attributes greater than 63 %. After analyzing the subsets, it can be seen that the difference of the AUPRC measure obtained in the subsets for the base with all the attributes is less than 0.025.