[Biomod-commits] Help on variable importance

Pablo Riul pabloriul at yahoo.com.br
Thu Nov 22 12:38:44 CET 2012

Dear Damien,

This is a good solution, thank you! 

Pablo Riul

Departamento de Engenharia e Meio Ambiente
Centro de Ciências Aplicadas e Educação
Universidade Federal da Paraíba - Campus IV
R: Mangueira s/n, Centro CEP: 58.297-000
Rio Tinto - Paraíba - Brasil

 De: Damien Georges <damien.georges2 at gmail.com>
Para: Pablo Riul <pabloriul at yahoo.com.br> 
Cc: "biomod-commits at lists.r-forge.r-project.org" <biomod-commits at lists.r-forge.r-project.org> 
Enviadas: Quinta-feira, 22 de Novembro de 2012 6:52
Assunto: Re: [Biomod-commits] Help on variable importance

Dear Pablo,

As you said, upper a variables importance score is, more effect
      the variable have (for the studied model).
The value returned is dependent from considered algorithm. I am
      not aware of any kind of consensus threshold to determine if a
      variable must be considered as important or not.. And I think it
      would not be fair to do it.

I guess the best, and the safest, way to interpret variables
      importance is to consider them relatively.. You can rank your
      variables by scores and say : for this model, this variable is
      more important than this other...



On 21/11/2012 21:31, Pablo Riul wrote:

Dear list,
>Is there any threshold to consider a predictor as an important variable? I understand the importance is the result of one minus the correlation between the original prediction and the prediction with the permuted variable. And higher the value higher the importance. But an reference value would be useful. Does anybody have any suggestion?
>Pablo Riul
>Departamento de Engenharia e Meio Ambiente
>Centro de Ciências Aplicadas e Educação
>Universidade Federal da Paraíba - Campus IV
>R: Mangueira s/n, Centro CEP: 58.297-000
>Rio Tinto - Paraíba - Brasil
Biomod-commits mailing list Biomod-commits at lists.r-forge.r-project.org https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/biomod-commits
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/biomod-commits/attachments/20121122/b1f8759c/attachment.html>

More information about the Biomod-commits mailing list