Métodos clássicos e bayesianos de estimação da janela ótima em núcleo- estimadores

Abstract

In this dissertation the estimation of univariate density functions through kernel methodology is discussed considering fixed and variable kernel estimators. The main purpose was to compare the performance of some methodologies of data-based bandwidth selection considering different approach of kernel estimation. Some methods for bandwidth selection under classics and bayesian approach were implemented. For fixed bandwidth only Plug-in (CHIU, 1991; SHEATHER; JONES, 1991) methodology was considered. For variable bandwidth classical and bayesian (BREWER, 2000; GANGOPADHYAY; CHEUNG, 2002) methods were evaluated. The comparison was performed by using Monte Carlo simulation. Different scenarios were simulated with the purpose to identify the advantages and peculiar characteristics of each methodology in terms of error criteria. Some examples of application were presented to show how the methodologies can be used in practical context.