Yes, I’ve been painstakingly training positive and negative cases for weeks. This is a standard TBird setup on imap with the adaptive filter enabled. Here’s the results from a 24 hour experiment:
- 96 spam incorrectly delivered to inbox (manually marked as spam)
- 257 messages automatically delivered to spam folder
- 3 of the above incorrectly (manually marked as not-spam)
Is this typical performance, or has something gone bad? Sifting through ~100 spammy messages a day is bad; losing 3 important things a day is worse. -m
As for the false negatives, spammers are getting much more clever. They’re using tricks based on MIME multi-part that are very hard to defeat without a high number of false positives. As for the false positives, I don’t think 3 out of almost 400 is bad at all. I think it’s typical, and it’s the price you pay for the fact that spam-filtering is an inexact science. Sometimes I get more false positives, sometims less. Not much to be done about that.