Análise de dados amostrais complexos utilizando redes neurais

• Ano de defesa: 2017
• Autor(a) principal: Pereira, Savano Sousa
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Rio de Janeiro; Brasil; Instituto Alberto Luiz Coimbra de Pós-Graduação e Pesquisa de Engenharia; Programa de Pós-Graduação em Engenharia Elétrica; UFRJ
• Palavras-chave em Português: Engenharia elétrica; Redes Neurais Artificiais; Abordagem de superpopulação; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: http://hdl.handle.net/11422/6335

The fitting of an Artificial Neural Network (ANN) considers the data coming from an simple random sample with replacement. However, in practice the selection of simple random samples for surveys is rarely used and more complex sampling schemes are used. The complex sampling schemes reflect complex structures from population. These structures of sampling scheme need to be incorporated when we fitting an ANN. In statistical literature, there are different aproaches for modelling data from complex surveys. However in the literature related to ANN there is no mention of how proceed when the data come from complex sample surveys. This work porpose an superpopulation approach for modelling data from complex survey using ANN. An evaluation of the proposed methodology is carried out empirically through simulations and, after, estimation measures are calculated. The practical application is done using data from the National Household Sample Survey for the year 2014. The objective is to improve the estimate of the per capita household income used to construct the Poverty Map in IBGE (2008).