characteristics of spam filterin

characteristics of spam filtering methods

Method	false Neg.	false Pos. 1	false Pos. 2	remarks
conservative DNSBL (against top known spammers)	15%	0.1%	0.01%
aggressive DNSBL (against potential spammers)	7%	2%~10%	1%~3%
Banning Servers from Policy-Based Netblocks (avoid all zombies)	25%	0.1%	0.5%
content keyword triggers	3%~15%	1%~3%	1%~3%
content keyword Bayesian pattern filter	2%~10%	0.5%~2%	0.5%~2%
content type ban policies (scripts, forms, etc.)	99%	15%	0.1%	mainly affecting automated email, such as order status notification; and newsletters.
Challenge-Response White-List human test	0.5%	100%	90%

Points to remember:

When combining methods, the resultant false negative ratio is subtractive, while false positive ratio is additive.
The damage of 1 piece of lost mail due to false positive equates that of 100 pieces of slip-thru due to false negative!