Prev: code
Next: Custom PDE equation
From: Eli on 11 Feb 2010 07:57
Hi,

I need to solve a classification problem and I would like to use a stepwise approach - add and remove features until I find an optimal set.

I tried using stepwisefit() from the statistics toolbox and it works ok but the results I'm getting are not so good since this optimization is based on regression rather than classification.

I would like to have a similar tool adapted for classification problems.
Any suggestions will be welcomed.

Rachel
From: Tom Lane on 11 Feb 2010 10:22
> I need to solve a classification problem and I would like to use a
> stepwise approach - add and remove features until I find an optimal set.
>
> I tried using stepwisefit() from the statistics toolbox and it works ok
> but the results I'm getting are not so good since this optimization is
> based on regression rather than classification.

Rachel, try "help sequentialfs" and look at the example at the end. It shows
sequential feature selection for classification using discriminant analysis.
If you have another classification method in mind, you may be able to adapt
the example to use your classifier.

-- Tom


From: Eli on 14 Feb 2010 00:37
Hi Tom, thank you for your reply.

I actually tried that method - but the problem is sequentialfs only selects 'forward' or 'backward', not both at the time, as I would like it to.

Do you have any other idea?

Thanks,
Rachel



"Tom Lane" <tlane(a)mathworks.com> wrote in message <hl17cb$nmu$1(a)fred.mathworks.com>...
> > I need to solve a classification problem and I would like to use a
> > stepwise approach - add and remove features until I find an optimal set.
> >
> > I tried using stepwisefit() from the statistics toolbox and it works ok
> > but the results I'm getting are not so good since this optimization is
> > based on regression rather than classification.
>
> Rachel, try "help sequentialfs" and look at the example at the end. It shows
> sequential feature selection for classification using discriminant analysis.
> If you have another classification method in mind, you may be able to adapt
> the example to use your classifier.
>
> -- Tom
>
 | 
Pages: 1
Prev: code
Next: Custom PDE equation