Is it true that Multinomial Naive Bayes requires equally by count training data for each class to get best performance?

For example, we forming classifier for three classes - Japan, China, Korea.

For Japan available 500 training data, China - 300, Korea - 100.

So to train Multinomial Naive Bayes we need to get only 100 data for each class inspite of we have 500 for Japan, 300 for China.

I know that we can train Naive Bayes with all available data, but in this case we got worse performance, is it true?

  • $\begingroup$ I found answer here aaai.org/Papers/ICML/2003/ICML03-081.pdf ! $\endgroup$ – nub Mar 1 '13 at 10:14
  • $\begingroup$ But I get another question: Is it the same for the classifiers like SVM? $\endgroup$ – nub Mar 1 '13 at 10:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.