[prev in list] [next in list] [prev in thread] [next in thread] 

List:       wekalist
Subject:    Re: [Wekalist] About Converting the Class for Linear Regression
From:       Paul <paul.m.nz () gmail ! com>
Date:       2006-04-26 20:03:23
Message-ID: a5ed8b040604261303x3210e3f5ta3c6426439c592ee () mail ! gmail ! com
[Download RAW message or body]

The error rate may also depend on the class spread (ie, what % of
instances belong to each class)... If it is imbalanced maybe a cost
sensitive classifier may do the trick...

For the multi class problem - how are you calculating the 25% error
rate (are you averaging the f-measures?)

/paul

On 4/26/06, Peter Reutemann <fracpete@waikato.ac.nz> wrote:
> > * Since my target class is nominal, would there be any serious side-effect
> > if
> > I converted it to a numerical type (i.e. D=0, I=1, A=2, B=3, C=4) and
> > applied
> > new algorithms to it ... to see if I can improve the error rate ?
>
> Well, I don't think so that it will make much sense... What will you do
> with a prediction of 1.6748 that you could get for an (unknown) instance?
> If you wanna play around with the original dataset but use regression
> schemes, try using the meta classifier "ClassificationViaRegression".
>
> > * I have tried some "binarizing filters" as pre-processing and this has not
> > improved the situation either.  Would this be equivalent to transforming my
> > target
> > class (nominal-to-numeric) as described in the previous paragraph ?
>
> I assume that the class attribute didn't get changed when you applied
> the binarize filters (normally, the class attribute is skipped). The
> binarization affected most likely only the other nominal attributes.
>
> HTH
>
> Cheers, Peter
> --
> Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
> http://www.cs.waikato.ac.nz/~fracpete/     +64 (7) 838-4466 Ext. 5174
>
> _______________________________________________
> Wekalist mailing list
> Wekalist@list.scms.waikato.ac.nz
> https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
>

_______________________________________________
Wekalist mailing list
Wekalist@list.scms.waikato.ac.nz
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist

[prev in list] [next in list] [prev in thread] [next in thread]