Vernon Schryver <vjs(_at_)calcite(_dot_)rhyolite(_dot_)com> wrote:
As others have pointed out, there is a genuine technical ambiguity in
"false positive." Is it RD/TOTAL or RD/SPAM? (for RD=rejected but
desired by recipient, TOTAL=total mail of all sorts whether rejected
or not, and SPAM=whatever that means).
As an absolute number, false positive should be easy to define. "I
went through my 'spam' folder, and I decided that 3 messages should
not have been put there."
As a ratio or percentage, false positives are taken relative to the
desired signal, not to the total data, or to the noise. So we have:
MAIL = number of emails you want to be marked as ok (not spam)
JUNK = number of emails you want to be marked as spam
MARKED = number of emails actually marked as ok
SPAM = number of emails actually marked as spam
FP = false positives (number of "MAIL" in "SPAM", which should
have been in "MARKED")
FN = false negatives (number of "MAIL" in "MARKED", which should
have been in "SPAM")
The ratio of false positives = FP / MAIL
The ratio o false negatives = FN / MAIL
MARKED = MAIL - FP + FN
SPAM = JUNK + FP - FN
Alan DeKok.
_______________________________________________
Asrg mailing list
Asrg(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/asrg
I wonder if it might be useful to adopt the widely used ideas of
specificity and sensitivity.
so if
TP = number of emails you want to be marked as spam that actually are so
marked
and
FP = number of emails marked as spam that shouldn't have been
and
FN = number of emails not marked as spam that should have been
specificity = TP/(TP + FP)
sensitivity = TP/(TP + FN)
(I think I've got that right - I'm sure there are lot's of references out
there)
These measures are commonly used for comparing diagnostic tests in clinical
applications.
--
_______________________________________________
Asrg mailing list
Asrg(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/asrg