View Full Version : Overview of weka classifiers for a three-class problem

03-03-2007, 12:40 PM
Hi peeps,
does anyone know which weka classifiers are able to work on a three-class problem?
Or do I have to find that on a trial-and-error basis?

03-09-2007, 07:34 PM
I would say it depends on problem in hands (I assume it's supervised learning). For getting around multi-class attribute issue, this question came up on original Wekalist. Here is the answer:
" If you have a dataset with multiple class attributes, and you want to process it with Weka, it needs to be split into several datasets, one for each class attribute."

For each data set you would train a classifier to pick one identified class and treat rest as other

check out the original thread:

03-13-2007, 06:22 PM

I'm assuming that the original question refers to the number of class labels (for a single class attribute) and not to the situation where there is more than one class attribute of interest.

Most Weka classifiers can handle multi-class problems (i.e. more than two class labels). Some handle this naturally. Other learners that are binary class learners achieve this via learning a classifier for each label (usually in a one-against-the-rest or pairwise fashion). If a given classifier does not do this directly, then it can always be wrapped up in a weka.classifiers.meta.MultiClassClassifier to achieve the same resutlt.