Learning Imbalanced Data

dragonedison

Dear everyone,
I would like to know if there is any learning operator in RapidMiner(either supervised or unsupervised) that is suitable for imbalanced data learning.

Thanks,
Gary

land

Hi,
what exactly do you mean by "suitable"? All learning algorithms can cope with imbalanced data, some better some worse. But many support weighting of examples so that you can even the odds. There are also many sampling operators, that might help to train on an imbalanced data set.

Greetings,
  Sebastian

dragoljub

I am also running into this problem, however I have had limited success with weighting the fewer example class with larger importance.

I find that sub-sampling the larger class generally helps.

I am trying to use LibSVM with the class weights, however they do not seem to do what is expected. How is the class weighting implemented in RM for LibSVM? It does not seem to be a standard option in the LibSVM C package?

Thanks,
-Gagi