Optimal Projections for Classification with Naive Bayes