Date Added: Sep 2009
Spam filters that are implemented using Na?ve Bayesian learning techniques are widely deployed worldwide with email clients such as Outlook. These filters that are deployed on end user's computers and typically used to filter out spam for individual users are effective when the spam load is around 400-500 spam emails per day per user. However, when the spam load increases, these solutions prove to be slow and hence insufficient for practical use. This paper identifies the computation intensive functions of such machine learning algorithms and solves the performance issues by implementing these functions on hardware.