Mahdi, NB assumes feature independence and there are many semi-NB classifiers that alleviate the attribute interdependence problem of NB. See e.g. Zheng and Webb. Lazy learning of Bayesian rules. Machine Learning, 41 (1):53{84, 2000.
and
Zheng and Webb. A comparative study of seminaive bayes methods in classification learning. In Proceedings of the Fourth Australasian Data Mining Conference (AusDM05), pages 141-156, Sydney, 2005.
3rd Mar, 2013
Mahdi Manafzade
Ferdowsi University Of Mashhad
Thanks dudes but I mean which is correct in the original NB "conditionally independent or independent features". I think features must be conditionally independent and can be dependent unconditionally.
3rd Mar, 2013
Pekka Jounela
Natural Resources Institute Finland (Luke)
Precisely it is conditionally independent; see http://en.wikipedia.org/wiki/Conditional_independence