Modelo de regresión lineal usando una mixtura de distribuciones senh-normal
Date
2023-09-27
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Pontificia Universidad Católica del Perú
Abstract
La distribución Senohiperbólico-Normal, denominada también como una variación de la distribución Birnbaum-Saunders, surgió inicialmente para estimar el deterioro en la calidad de los materiales sujetos a estrés. Asimismo, los modelos de mixtura han suscitado considerable interés en el campo de estadística debido a que permiten lidiar con situaciones en las que el comportamiento de los errores de un modelo con ajuste lineal se aleja significativamente de la normalidad. Esta tesis aborda los dos temas mencionados mediante la presentación de un modelo de ajuste lineal usando una mixtura de distribuciones Senohiperbólico Normal o Log-Birnbaum-Saunders. Esta propuesta es una familia versátil de distribuciones de probabilidad que posibilita representar datos que presentan multimodalidad además de provenir de poblaciones heterogéneas. Para conseguir los estimadores de máxima verosimilitud se emplea el algoritmo EM con maximización condicional. Asimismo, se llevan a cabo estudios de simulación y análisis de conjuntos de datos reales para demostrar la utilidad del método propuesto. Por último, se implementa la propuesta del algoritmo y los métodos en el software R.
The Senohyperbolic-Normal distribution, also known as a variation of the Birnbaum-Saunders distribution, was initially developed to estimate the deterioration in the quality of materials subjected to stress. Likewise, mixture models have attracted considerable interest in the field of statistics because they allow dealing with situations where the error behavior of a linearly fitted model deviates significantly from normality. This thesis addresses the two aforementioned issues by presenting a linear fitting model using a mixture of Senohyperbolic Normal or Log-Birnbaum-Saunders distributions. This proposal is a versatile family of probability distributions that makes it possible to represent data that present multimodality as well as coming from heterogeneous populations. The EM algorithm with conditional maximization is used to obtain the maximum likelihood estimators. Simulation studies and analysis of real data sets are also carried out to demonstrate the usefulness of the proposed method. Finally, the proposed algorithm and methods are implemented in R software.
The Senohyperbolic-Normal distribution, also known as a variation of the Birnbaum-Saunders distribution, was initially developed to estimate the deterioration in the quality of materials subjected to stress. Likewise, mixture models have attracted considerable interest in the field of statistics because they allow dealing with situations where the error behavior of a linearly fitted model deviates significantly from normality. This thesis addresses the two aforementioned issues by presenting a linear fitting model using a mixture of Senohyperbolic Normal or Log-Birnbaum-Saunders distributions. This proposal is a versatile family of probability distributions that makes it possible to represent data that present multimodality as well as coming from heterogeneous populations. The EM algorithm with conditional maximization is used to obtain the maximum likelihood estimators. Simulation studies and analysis of real data sets are also carried out to demonstrate the usefulness of the proposed method. Finally, the proposed algorithm and methods are implemented in R software.
Description
Keywords
Análisis de regresión, Teoría de las distribuciones (Análisis funcional), Algoritmos
Citation
Collections
Endorsement
Review
Supplemented By
Referenced By
Creative Commons license
Except where otherwised noted, this item's license is described as info:eu-repo/semantics/openAccess