Open set diagnosis: high-dimensional clustering

It is customary to discriminate between a finite number of a complex system' states and to assign a new observation to the one with the highest posterior probability. Nevertheless, it is not actually obvious to correctly describe all of the possible states, neither to identify their exact number, some might be rare, some too risky or costly to obtain. Hence, it is vital to study how to affect new observations that deviates considerably from the closed training data. In this paper, a new open set decision scheme clustering-based has been proposed to deal with high-dimensional, unlabelled observations and undefined states.