Critérios de seleção de modelos: um estudo comparativo

• Ano de defesa: 2021
• Autor(a) principal: Moura, Adriana Ribeiro
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Informática; Programa de Pós-Graduação em Modelagem Matemática e computacional; UFPB
• Palavras-chave em Português: Seleção de modelos; Critérios de informação; Distribuições Sup e Inf; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/23007

Several scienti c researches in various areas, including Statistics, have their study problems linked to practical situations, which can usually be explained through models, and it is common for researchers to come across more than one model describing the same phenomenon. Given this fact, authors defend the need for a standard criterion based on scienti c principles for choosing the model that best explains the phenomenon and the literature already has several criteria for selecting models with this objective. Thus, considering the families of generalized distributions Sup and Inf, the present work aims to propose a new model selection criterion for non-embedded models, based on these families of distributions and their properties and to compare their performance with the criteria: information criterion of Akaike (AIC), corrected Akaike information criterion (AICc), Bayesian information criterion (BIC), Hannan information criterion- Quinn (HQIC) and the modi ed t criteria of Crámer-Von Mises (W ) and Anderson-Darling (A ) through di erent simulation scenarios. Also for comparison purposes, its applicability was illustrated using real datasets. Additionally, the most important results on multiple linear regression were presented and simulations were performed in order to compare the performance of the new proposed criterion with the AIC, AICc, BIC and HQIC criteria in the selection of regression models, as well as an application to a set of real data.