When configuring a rule you can specify the Bayesian filter threshold that must be reached to trigger the rule. The higher the threshold, the more certain the Bayesian filter must be that the message is spam.