Naive Bayes: Subtleties

1. Conditional independence assumption is often violated
   
   109#109
   
   
   • ...but it works surprisingly well anyway. Note don't need estimated posteriors 125#125 to be correct; need only that
   
   
   126#126
   
   

   
   

   • see [Domingos & Pazzani, 1996] for analysis
   • Naive Bayes posteriors often unrealistically close to 1 or 0