Hello,
We currently get on average, with a 3,500 customer base, 291,293 emails
per day. Of those, we usually catch as spam around 75,000 per day. This
means we are only catching about 25.7% of incoming SPAM. It seems as though
everyone else is getting around 60% or higher. What is the number of false
positives you are seeing? Maybe you could send us your config files/custom
rules, because ours seem to be... lacking.
Our false positive rate is about 0.27%, with false negatives at 0.48%, for an overall efficiency of 99.24%. As others here have mentioned, this is attributed to a number of factors:
(1) A well-trained Bayes database. Mistakes (the false positives and false negatives) are corrected in batches on an hourly basis, with sa-learn. False positives and negatives are marked as such by users via the Maia Mailguard web interface. User-confirmed spam is also reported to Razor2, Pyzor, and DCC during this process.
(2) DNSBL tests (by SpamAssassin, not at the MTA stage).
(3) Collaborative reporting networks--Razor2, Pyzor, and DCC. We consult all three.
(4) SpamAssassin's default rules.
(5) Custom rulesets. We use the Bigevil, Evilnumbers, Backhair, and Weeds lists.
The external tests, particularly the DNSBLs, and the collaborative networks, make a huge difference, in our experience. They may "cost" more in terms of processing time and network latency, but they add a lot of value to the overall analysis. Doing a thorough analysis adds a couple of extra seconds to the processing of each mail item, but it pays off in terms of improving the efficiency of the filter.
Robert LeBlanc <[EMAIL PROTECTED]>
Renaissoft, Inc.
Maia Mailguard <http://www.renaissoft.com/maia/>
