Uso de métodos robustos para tratamento de outliers na estimação de parâmetros do modelo logístico

• Ano de defesa: 2025
• Autor(a) principal: Manoel, Iuri dos Santos
• Outros Autores: https://orcid.org/0000-0002-2018-309X
• Orientador(a): Fernandes, Tales Jesus
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Lavras; Instituto de Ciências Exatas e Tecnológicas (ICET)
• Programa de Pós-Graduação: Programa de Pós-Graduação em Estatística e Experimentação Agropecuária
• Departamento: Departamento de Estatística
• País: brasil
• Palavras-chave em Português: Métodos estatísticos; Reamostragem (Estatística); Modelos não lineares; Análise de regressão; Coelhos; Statistical methods; Resampling (Statistics); Nonlinear models; Regression analysis; Rabbits
• Área do conhecimento CNPq: Ciências Agrárias
• Link de acesso: https://repositorio.ufla.br/handle/1/60178

In data analysis, it is common to use statistical models to describe the relationship between the variables under study. However, when there are outliers in the sample set, the estimates of the model parameters may be biased. Therefore, the detection and appropriate treatment of outliers is essential to ensure representative results. In this context, robust methods can be a solution for dealing with outliers, as theys can mitigate the effects that these values have on estimates of interest. Among the diversity of robust methodologies disseminated in the literature, the robust methods Jackknife and Least Trimmed Squares (LTS) stand out. These methods are efficient and versatile in selecting sample subsets with fewer outliers than the sample set. This allows selecting a proportion of the sample that best represents it. In view of the above, this work aims to propose innovative robust methodologies applied to the description of growth curves. These methods are presented in three chapters in which the proposal is to obtain an adjustment of the nonlinear logistic model that best represents the data set. In the first chapter, the Jackknife resampling method is implemented by considering estimates for the kurtosis coefficient as a selection criterion. In the second chapter, the LTS method is implemented by considering sample partitions to select values according to the sum of squared deviations (SSD). It is worth noting that in these first two chapters the proposed methodologies are applied to the description of a set of simulated data via Monte Carlo. Finally, in the third chapter both methods are applied to the description of a set of real data. The data are from rabbit farming and come from the Department of Animal Science of the Federal University of Lavras. The observed measurements are the weight and length of the front paw of the rabbits. For the implementation of the methodologies, a breaking point is considered as a criterion to determine the size of the subsets selected for the adjustment. In addition, the comparison between adjustments is made by the mean absolute deviation (MAD). The methodologies proposed in this work are promising in adjusting the Logistic model in the description of growth curves, with the LTS method presenting better results than the Jackknife method. The break point and the MAD are appropriate as criteria to determine the size of the sample subset and comparison between the adjustments, respectively.