Bayesian mixture of parametric and nonparametric density estimation: A Misspecification Problem

In this paper we study the effect of model misspecifications for probability density func tion estimation. We use a mixture of a parametric and nonparametric density estima tion. The former can be modeled by any suitable parametric probability density function, including mixture of parametric models...

Descripción completa

Detalles Bibliográficos
Autores: Dias, Ronaldo, HEDIBERT FREITAS LOPES
Tipo de recurso: artículo
Estado:Versión publicada
Fecha de publicación:2011
País:Brasil
Institución:Instituição de Ensino Superior e de Pesquisa (INSPER)
Repositorio:Repositório Institucional da INSPER
Idioma:inglés
OAI Identifier:oai:repositorio.insper.edu.br:11224/4042
Acceso en línea:https://repositorio.insper.edu.br/handle/11224/4042
https://doi.org/10.12660/bre.v31n12011.4134
Access Level:acceso abierto
Palabra clave:Nonparametric Density Estimation
B-Splines
Mixtures Models
MCMC
EM-Algortihm
Descripción
Sumario:In this paper we study the effect of model misspecifications for probability density func tion estimation. We use a mixture of a parametric and nonparametric density estima tion. The former can be modeled by any suitable parametric probability density function, including mixture of parametric models. The latter is given by the known B-spline es timation. The procedure also deals with the situation when a highly structured data are collected so that it is difficult to propose a parametric model with a large number of mixture components. Then a nonparametric part would help to postulate an appropriate model. In addition, in order to reduce the computational cost of getting a nonparamet ric density for high dimensional data a parametric mixture of densities could be used as the starting point for modeling such dataset. Our procedure is computed by using EM-type algorithm for a non-Bayesian approach and MCMC algorithm under a Bayesian point of view. Simulations and real data analysis show that our proposed procedure have performed quite well even for non structured datasets.