# Spam Probabilities

Anti-spam rule files contain sets of test definitions that are used to identify potential spam features in an email message. Each feature, regardless of the rules file it is defined in, has an associated weight that contributes to the message's total spam score.

Weights are specified as numerical values and can be either positive or negative numbers. Positive weights increase the likelihood that a message is spam, while negative values decrease the likelihood.

An email message with several positive spam features will thus have a higher aggregated score than a message with negative or few spam features. Generally, a message must contain multiple spam features in order to result in a high aggregated spam score.

After an email message is scanned and all features are tested for, the anti-spam engine totals all associated weights into a spam score. The engine then converts the spam score into a percentage indicating the probability that the message is spam.

## Example: A Spam Percentage

`Spam probability for this message: 99.412%`

The following formula shows how the spam score is converted into a spam probability:

`PROB = 1/(1 + exp(-(BIAS+SCORE)/2))`

Where `BIAS`

= -5 and `SCORE`

is the sum of weights of triggered
features.

The chart below shows the relationship between scores and percentages. A score of '5' results in a 50% spam probability.