the same for everyone, but I want to get the feel of general
statistics (If you don't mind to share)
1. How many Spam detection rate if I am using default 3.2
configuration you would expect?
> 2. If fine tuned according to the wiki, e.g. running sa-update, more
> rules set, how many % you would expect then?
Every calculation depends on how many spam you already reject at MTA
level. If you pass everything on to SA, you have a higher spam:ham ratio
than if you can reject the trivial spam at MTA level and pass on only
the difficult-to-detect ones.
I am running a tiny (5 users) Postfix system that rejects many malformed
delivery attempts as well as unknown sender or recipient. It also does
greylisting and rbl checking (dsbl, spamhaus). Approximatelys 30% of all
mail is accepted and given to SA. From 100 mails, I get ~1 false
negative, that yields 99% accuracy with spam detection. It is many
months ago I got the last false positive, so I would say 0.01% accuracy
with false positives.
This configuration is already very well tuned with Bayes-learning, ZMI
rules for german spam, sought rules, and of course DCC, razor and pyzor.
sa-update is run once a day.
For the default ruleset I guess an accuracy of perhaps 95-97% accuracy
and same false positive rate as above.
3. Is the % vary from SA version? e.g. 3.0, 3.1 and 3.2?
Older versions yield significant lower accuracy, since the spam
structure changes every week and the code of SA is modified constantly
to accomodate this. In many cases, simple rule changes are not
sufficiant to catch up.
Tschau
Alex
