Otimização multiobjetivo para seleção simultânea de variáveis e objetos em cromossomo duplo de representação inteira para calibração multivariada

• Ano de defesa: 2017
• Autor(a) principal: Bastos, Hélios Kárum de Oliveira
• Orientador(a): Soares, Anderson da Silva
• Banca de defesa: Soares, Anderson da Silva, Sanches, Danilo Sipoli, Laureano, Gustavo Teodoro
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Goiás
• Programa de Pós-Graduação: Programa de Pós-graduação em Ciência da Computação (INF)
• Departamento: Instituto de Informática - INF (RG)
• País: Brasil
• Palavras-chave em Português: Calibração multivariada; Algoritmo genético; Otimização multiobjetivo; NSGA-II; Representação inteira
• Palavras-chave em Inglês: Multivariate calibration; Genetic algorithm; Multiobjective optimization; NSGA-II; Integer representation
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://repositorio.bc.ufg.br/tede/handle/tede/8094

Multiobjective Optimization for feature and samples selection in double chromosome of integer representation and variable size for multivariate calibration} In several problems of regression, classification, prediction, approximation Optimization, the original data contain a large number of variables to obtain a better representation of the problem under consideration. However, a significant part of the variables may be irrelevant and redundant from the point of view of machine learning. Indeed, one of the challenges to be overcome is a selection of a subset of variables that has the best perform. One of the breakthroughs in this type of problem is the use of a multiobjective formulation that avoids the overlap of the model to the training data set. Another important point is the process of choosing the objects to be used in the learning stage. Generally, a selection of variables and treatment objects are treated separately and without dependence. This project proposes a multiobjective modeling to select variables and objects simultaneously using a genetic integer representation algorithm with variable size chromosomes. It is expected that a simultaneous selection of objects and variables on a multiobjective context produce better results in a traditional approach. As a case study this work utilized an analysis of near infrared (NIR) material on oil samples for the purpose of estimating the concentration of an interest properties such set was used in the competition conducted at the International Diffuse Reflectance Conference (IDRC) in the year 2014.