Negative merit values obtained when using the ClassifierAttributeEval

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Negative merit values obtained when using the ClassifierAttributeEval

MiguelHK
Hello,
I am trying to perform attribute (genes) selection with the ClassifierAttributeEval, choosing accuracy as the evaluation measure and Random Forest as the classifier.
My data has the expression levels (numerical values) of 13677 genes for 27 instances (samples), 6 pertaining to one class and 21 to a second class (the class is nominal). There are no missing values, NAs, etc..

As far as I understand the ClassifierAttributeEval will try to build Random Forests gene by gene and then rank them based on the accuracy. Therefore, the merit column should be the average accuracy of the models built (I am using 5 fold cross-validation). However, I am obtaining negative values as the average merit for some genes. How is that possible?

Thanks in advance,
Miguel
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to: To unsubscribe send an email to [hidden email]
To subscribe, unsubscribe, etc., visit
https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|

Re: Negative merit values obtained when using the ClassifierAttributeEval

Eibe Frank
There are some fairly recent relevant posts in the archive:


Let us know if these explanations are insufficient.

Cheers,
Eibel

On Wed, Feb 12, 2020 at 12:31 AM <[hidden email]> wrote:
Hello,
I am trying to perform attribute (genes) selection with the ClassifierAttributeEval, choosing accuracy as the evaluation measure and Random Forest as the classifier.
My data has the expression levels (numerical values) of 13677 genes for 27 instances (samples), 6 pertaining to one class and 21 to a second class (the class is nominal). There are no missing values, NAs, etc..

As far as I understand the ClassifierAttributeEval will try to build Random Forests gene by gene and then rank them based on the accuracy. Therefore, the merit column should be the average accuracy of the models built (I am using 5 fold cross-validation). However, I am obtaining negative values as the average merit for some genes. How is that possible?

Thanks in advance,
Miguel
_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to: To unsubscribe send an email to [hidden email]
To subscribe, unsubscribe, etc., visit
https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html

_______________________________________________
Wekalist mailing list -- [hidden email]
Send posts to: To unsubscribe send an email to [hidden email]
To subscribe, unsubscribe, etc., visit
https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html