In this paper, a set of simple predictive criteria, each optimized for a given type of disruption, is explored. Disruptions that occurred in the years from 2005 to 2009 in the ASDEX Upgrade tokamak are classified into several types in a first step. Then, discriminant analysis is used as the main approach to the disruption prediction and a log-linear discriminant function, constructed with five global plasma parameters that have been selected from an initial group of ten variables, is derived for the edge cooling disruptions. The function is tested off-line over 308 discharges and is shown to work reliably. It describes a clear dependence of the disruption boundary on the plasma parameters.