thresholds and accuracy (Re: FW: Why is this spam passing my SA (counterfeit goods))
Jonas Eckerman writes:[color=blue]
> Josie Walls wrote:[color=green]
> > Would this group agree that requiring 5 hits in order to classify an email
> > as spam is too conservative a number?[/color]
> I wouldn't.
> We reject everything with a score of 18 or higher, quarantine
> everything with 9 or higher (I read the quarantine reports every
> work day), and let everything else get through with spam score
> (numeric and as "**") in headers.
> Individual users are free to filter based on the headers if they
> like, but most don't.
> Of course, the optimal thresholds depends on how your system is
> setup, what happens when the score gets over a threshold, what
> rules you use, how well your bayes system works, how important it
> is to avoid FPs etc, etc.[/color]
Sounds like a great opportunity to mention this chart I made from
the thresholds used in SpamAssassin 3.2.0: