hi,
I have 5000 data exemplars with 3 classes... but distribution of the classes (1 , 2, 3) are about 50/30/20 ... classification of 2 and 3 is rather poor compared to 1.. to achieve better classification results for them I used MetaCost for weighting class 2 and 3 in the testing / prediction process....
but I would also like to do weighting of the classes 2 or 3 in the learning process of the model... before testing, how can I do that? is there a way to scale up my underrepresented classes with an operator? or what are good weighting functions for learning models?