Spam Bayes - Very High False Positives

[CeleronXT]

I've been using SpamBayes for MS Outlook ever since PCWorld highly recommended it. Many people in MozillaZine's AfterDark section absolutely rave about Bayes. I find that it does an excellent job at catching spam, and spam never gets through to the Inbox. Ever. However, it also has a very high false-positives count. Between 40%-60% of all the mail from my site goes straight to the suspected spam bin (not to be confused with the definite spam bin). Good mail very rarely goes to the definite spam bin, but almost everything that goes to the suspected bin is good mail.

I've trained and retrained it for good/bad mail, but this problem is constant. Has anyone else had such a problem with it and been able to fix it?

[Freddie Bingham]

When that happens I just click "recover from spam" and it 'learns' not to add that mail any more.

[Mike Sullivan]

At the beginning I had that, particularly from emails with attachments, but now it works very well.

[chrispadfield]

This is my problem with these spam systems, they make false postivies. I can live with false negatives but not false postives. Cloudmark also has given very high postivies, particularly any email that is autogenerated (and someone thus can't be bothered to subscribe to).

[CeleronXT]

I do the recover, but I still constantly get false positives. I've been using it for quite a while, and it's been in training the whole time (well obviously..), and its still getting them. PCWorld said that after training for just about a week they got it working nigh' perfectly.

Most emails with an attachment still go to the bin, as do most forum registration confirmation messages.

Why does anyone send spam anyway? No one who gets spam actually buys, do they?

:/

Edit - My avatar looks crappy in Internet Exploiter with the crappy PNG support.

[Raz Meister]

Install SpamAssassin, it's much better: http://www.vbulletin.com/forum/showp...6&postcount=56

[CeleronXT]

This is my ISP's email, so I don't have access to the server.

[Steve Machol]

I've had great success with SpamAssassin. Out of thousands and thousands of spam emails over the last couple of weeks - it catches about 98% of them, and I've only had two false positives - neither of them important.

[Freddie Bingham]

I don't know, Spambayes works pretty damn good for me. I get a couple "suspected" emails every day and rarely is it a false positive.

[jw791]

I just installed it yesterday, and it seemed to work. 1 false positive and 5-6 correct hits with little training

[CeleronXT]

I just decided to try something different. I retrained it without emptying the database, and set the ordinary good folders, but set an empty folder for bad mail. Maybe this will raise percent values on good words while leaving the bad ones as-is. I'm hoping the problem is something like that. :/

Every one of the slashdot mailers goes to the suspected bin. I recover it every time every night, and it still goes there.

[Freddie Bingham]

Maybe it is on to something when it suspects the slashdot emails

[CeleronXT]

It didn't trash the slashdot mailer this time. Looks like it's improving a bit. Although it looks like Freddie would think this was actually a regression in performance.