2 Swlpl ] Lqj Wkh ( Uuru 5 Hmhfw 7 Udgh Rii Iru D 0 Xowl ( [ Shuw 6 \ Vwhp Xvlqj Wkh % D \ Hvldq & Rpelqlqj 5 Xoh

$EVWUDFW Recently, in the framework of Pattern Recognition, methods for combining several experts (Multi-Expert Systems, MES) in order to improve the recognition performance, have been widely investigated. A main problem of MES is that the combining rule should be able to take the right classification decision even when the experts disagree. Anyway, in critical cases, a reject decision is convenient to reduce the risk of an error. Up to now, the problem of defining a reject rule for a MES has not been systematically explored. We propose a method for determining the best trade-off between error rate and reject rate depending on the considered application domain, i.e. by taking into account the costs attributed, for the specific application, to misclassifications, rejects and correct classifications. Even though the method has general validity, in this paper its application to a MES using the Bayesian combining rule is presented.