On 17/06/12 2:56 AM, claudia87 wrote:
>
> Hi,
>
> I'm using WEKA to classify in 3 classes. On a few of my input files it
> scores 0 for one class. This is not one type of class put for each input it
> chooses a different class.
>
> For example I get this output:
>
> === Detailed Accuracy By Class ===
>
> TP Rate FP Rate Precision Recall F-Measure ROC Area
> Class
> 0 0 0 0 0 0.656
> pos
> 0.117 0.04 0.676 0.117 0.2 0.733
> neg
> 0.973 0.892 0.471 0.973 0.634 0.776
> neutr
> Weighted Avg. 0.485 0.417 0.491 0.485 0.367 0.742
>
> === Confusion Matrix ===
>
> a b c <-- classified as
> 0 11 125 | a = pos
> 0 48 361 | b = neg
> 0 12 432 | c = neutr
>
>
> It works fine with the same input on another classifying program and when I
> use the InfoGain filter to reduce the attributes it also works fine.
>
> I can't find a similar problem on Google anywhere. Does anybody have any
> idea what the cause of this can be.
>
>
> Edit: Maybe some information that is of use. The input file is a the input
> of 2 files. So it gets all the attributes of the 2 files, but only the
> instances of 1, so I can use the 2 files together as seperate test and
> training set, else it says that files are not similar. So not all attributes
> that are in the attribute set are there in the instances.
>

Given the default threshold of 0.33 on the probability of choosing one
class over another on this problem, it's the case that the classifier
never predicts "pos". However, the ranking performance for the "pos"
class is still reasonable (as shown by the ROC area).

Cheers,
Mark.

_______________________________________________
Wekalist mailing list
Send posts to:
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
)