An algorithm combining multiple Nave Bayesian (NB) filters based on GMM is presented

which has been successfully applied to e-mail filtering.The method uses the multiple variates statistics analysis to model the relationship between the training data set and their classification by a collection of NB filters.Then a GMM can be learned from the resulting representation.The GMM filters previously unseen e-mails according to the principle of minimizing expected-error-cost

in order to avoid deleting useful e-mails.Experimental results confirm the validity of our method

and show that our approach is insensitive to ratio of feature subset selection.