[ABE-L] Seminário

[Caio Lucidius Naberezny Azevedo]

Prezados,


Boa noite. Tenho o prazer de anunciar (abaixo) o próximo seminário do ciclo
de palestras do programa de Pós-Graduação em Estatística do IMECC-Unicamp.


Todos estão convidados.


Cordialmente,


Caio Azevedo



Title: A robust regression method based on exponential-type kernel functions


Palestrante: Prof. Dr. Eufrásio de Andrade Lima Neto, Departamento de
Estatística, UFPB

Data: 25/08/2017, 14h00 sala 121, IMECC-Unicamp

Abstract: The use of robust regression methods is common in practical
situations due to the presence of outliers. This work proposes a robust
regression method that re-weighted the outliers observations considering
type-exponential kernel functions. The convergence of the parameter
estimate algorithm is guaranteed with a low computational cost. A
comparative study between the proposed regression method (ETKRR) against
some classical robust approaches and the OLS method is considered. We have
considered synthetic datasets with X-axis outliers, Y-axis outliers and
leverage points, in a Monte Carlo simulation framework with different
sample sizes and percentage of outliers. The results have demonstrated that
the ETKRR approach presented a competitive (or best) performance in
simulation scenarios that are similar to those found in real problems.
Applications to real datasets has showed the usefulness of the proposed
method.


-- 

===================================================
Assistant Professor,
Department of Statistics, Institute of Mathematics,
Statistics and Scientific Computing,
University of Campinas, Brazil.

Office Phone : 55 - 19 - 35216060
Fax : 55 - 19 - 3521-6094

home-page : http://www.ime.unicamp.br/~cnaber/
===================================================

"Pois desde a criação do mundo os atributos invisíveis de Deus, seu eterno
poder e sua natureza divina, têm sido observados claramente, podendo ser
compreendidos por intermédio de tudo o que foi criado, de maneira que tais
pessoas são indesculpáveis. "
Romanos 1:20
-------------- Próxima Parte ----------
Um anexo em HTML foi limpo...
URL: <https://lists.ime.usp.br/archives/abe/attachments/20170802/267a85d9/attachment.html>