From: "David G. Miller" <dave@xxxxxxxxxxxxx>
"jdow" <jdow@xxxxxxxxxxxxx> wrote:
> If you run your own mail server, I've had good luck with dSpam. It
> takes a little while to get it trained since it's Bayesian filter based
> but I now get overall accuracy of over 95% and the spam identification
> is over 93%. I should add that I get *very few* false positives
> (currently 12 out of over 15,000 e-mails) and several of those were
> right after installation when the filter had very little data to go on.
Wow - at THAT level it'd get chucked with extreme prejudice. (Since it
is not actively evil I'd not give in the paraffin and lead foil wraps
followed by a staking and burial at midnight of a full Moon. But I would
put it on a disk and bury the disk.) That accuracy is horrible. If I cut
my spam catch percentage to about 99% or so I could virtually eliminate
my false alarm rate. But that's not worth the effort tuning the system.
It ain't broke so I am not going to fix it. I'm serious that when I had
the 250-300 spam a day spam traffic I'd go one to two weeks without a
false alarm, once I found some juju to handle technical mailing lists
that have patch files cross the list frequently.
SpamAssassin can be awkward to setup. But once setup it is devastatingly
effective, especially since it scores block lists. No one block list
will give you a false positive. But they do save a lot of spam sneak-
throughs. I'm quite convinced that a proper anti-spam tool combines
rules and Bayes.
{^_^}
My guess is that the dSpam developers error on the side of fewer false
positives. The above results are with the sensitivity set as high as it
goes. dSpam is really designed to be forgiving to users like my wife.
She checks her quarantine about once every few weeks and then usually just
zaps everything that's in there. Her approach only works well if there
are very few false positives. It also allows me to have a server-side
spam filter instead of trying to set up some separate product for her
Windoze box.
Loren and I both use Outlook Express since that's the machinery we use
when we are making money. I just turned on DoveCot and Bob's my uncle.
If she's not complaining I'd leave hers as is and set mine up with
SpamAssassin. It gets to be a game to see how well I can keep it running
with near perfect statistics. I have made it for about three weeks without
a false anything - once. "I won!" (With DoveCot I can even log in and
fetch down my mail (ssl link of course) all neatly filtered when I am
away from home for conventions.)
Heh, real life is a better computer game than any game wizard can write.
(I've even had one of the "top of the line" spammers respond to a jibe
on the SA users list. I cracked wise about some of his spam scores not
quite making it to 100 points. Two days later in came the spam with his
"signature" traits in it with 100 points in small scores. He hit just an
amazing number of rules. The highest scoring rule he hit was the rule for
translating really high Bayes scores.)
{^_-} Simple pleasures for simple minds.
