Abstract
We study properties of the index J(3), defined as the accuracy, or the maximum correct classification, for a given three-class classification problem. Specifically, using J(3) one can assess the discrimination between the three distributions and obtain an optimal pair of cut-off points c(1)