Please use this identifier to cite or link to this item: http://repositorio.ufla.br/jspui/handle/1/11012
Title: Procedimento para identificar outliers por meio da distribuição acumulada de mínimo em um modelo com resposta gama
Other Titles: Procedure to identify outliers through the minimum cumulative distribution on a model with gamma response
Authors: Cirillo, Marcelo Ângelo
Muniz, Joel Augusto
Brighenti, Carla Regina Guimarães
Keywords: Simulação Monte Carlo
Distância de Mahalanobis
Taxa de mistura
Dispersãoparâmetro de dispersão
GLM
Monte Carlo simulation
Mahalanobis distance
Mixing rate
Dispersion parameter
Issue Date: 8-Apr-2016
Publisher: Universidade Federal de Lavras
Citation: RESENDE, M. Procedimento para identificar outliers por meio da distribuição acumulada de mínimo em um modelo com resposta gama. 2016. 50 p. Dissertação (Mestrado em Estatística e Experimentação Agropecuária)-Universidade Federal de Lavras, Lavras, 2016.
Abstract: This study aimed to propose a procedure based on the accumulated distribution of minimums to identify generalized outlier models by using Gamma response. To validate this methodology, we used Monte Carlo simulation, considering the scenarios defined by the combination of different sample sizes, the contamination rate and the distributions with different degrees of asymmetry. In this context, probabilities related to classification and accuracy errors were obtained from 500 Monte Carlo achievements. We concluded that the method is effective for presenting high accuracy probability. In terms of implementation, through the illustrated example, given the similarity between the new proposed approach, compared to approaches based by the lever matrix and Cook’s distance, we conclude that the procedure suggested in this study is feasible for implementation in response to Gamma distribution.
URI: http://repositorio.ufla.br/jspui/handle/1/11012
Appears in Collections:Estatística e Experimentação Agropecuária - Mestrado (Dissertações)



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.