what is the correlation between spam weight and probablility

Discussion, support and announcements for third party applications that work with MailEnable.
Post Reply
rdewild
Posts: 6
Joined: Wed Oct 24, 2012 3:20 pm

what is the correlation between spam weight and probablility

Post by rdewild »

We are using ME Enterprise v. 6.0 on Windows Server 2003 x64 SP2.
We have spam protection enabled and assign various "positive weightings" to the various spam detection options. (See Spam Protection in the ME control panel)
In the MTA filter report we can clearly see how messages are marked with these weights. e.g.

SMTP [System Spam Filter] ADD_HEADER [SMTP:bounce-64-25293-34255@some.domain.com] 123.123.123.123 High (235)

However, when we setup our filters and more specifically the filter criteria "Where the message has over a certain spam probability" the option to indicate is a "percentage" value. e.g. ME suggests (values over 90% recommended).

My question is, how does the "spam weighting" correlate to the spam probability as a percentage or do they? Am I not comparing apples to apples here?

poweredge
Posts: 157
Joined: Sat May 29, 2021 11:16 am

Re: what is the correlation between spam weight and probablility

Post by poweredge »

Just came across this topic while google

That's what I would like to know too

For example
SF [System Spam Filter] ADD_HEADER [SMTP:user@domain.com] xxx.xxx.xxx.xxx High (1030)

The score is 1030, and percentage is 90%,, these two are different

Update, found the answer in manual, the percentage applies to Bayesian filter only. It's better to update as "Where the message has over a certain spam probability (Bayesian filter)", or ppl will get confused this % with weighted spam score (ie, low/medium/high)

"
Where the message has over a certain spam probability

Filter to set the threshold for spam probability of Bayesian filtering e.g., define the filter to mark messages as junk if they have over 96.5% spam probability. See Bayesian filtering section for information on configuring the Bayesian filter."

Post Reply