[Spambayes] Spambayes vs. Popfile and other Bayesian classifiers

Peter Strisik lists at strisik.com
Wed Dec 3 13:07:07 EST 2003


Seth,

I don't keep any stats on performance, though I do know that I average about
30 spam per day.  By reliable, I mean consistent.  Outclass keeps working
with very occasional false negatives.  Virtually no false positives.  And it
did this without having to do much training.  With SpamBayes, I had to have
messages available, worry about how many spam and ham I was using, etc.  An
I found that it drifted towards poorer performance missing spam messages
requiring some retraining.  Basically, I just experienced outclass working
like I needed without effort or worry.

I don't use any white/black list features in outclass, only the bayesian
filtering.

......Peter


-----Original Message-----
From: spambayes-bounces at python.org [mailto:spambayes-bounces at python.org] On
Behalf Of Seth Goodman

Outclass does have some nice UI features, as does the parser in the
underlying PopFile.  Could you be more specific as to what you mean by "more
reliable"?  Though SpamBayes still produces a fair amount of false negatives
for my mail stream, I have yet to experience a false positive.  This is
extremely important, as I *never* want to lose an important email in a large
spam folder.  Visually inspecting the large spam folder is not totally
reliable.  Could you state approximately what were your false negative
rates, false positive rates and total spam load per day?  Since Outclass
(and PopFile) support whitelists and blacklists, how much use did you make
of those features?




More information about the Spambayes mailing list