In the field of pattern recognition, traditionally, the theory assumes the existence of two or more, presumably, well-defined, classes. The theory also assumes that the practitioner has a "sufficient" supply of descriptive data points from their respective classes. Recent research has shown, however, that there are classes of problems where this fundamental model does not hold. Under these circumstances, data from the second class is either absent severely limited, thus, dictating an alternative approach to training and classification.

This research endeavour aims to explores the classification capabilities of pattern recognition systems which are trained on a single class. In particular, domains characterized by some consistent background noise from which a set of rare, episodic events should be distinguished are of interest. The difficulty with such a classification is that the occurrence of these events is both random and unpredictable. In spirit of the Comprehensive Test Ban Treaty, we have undertaken the development of an application to simulate the emission of radionuclides from a series of industrial emitters along with the episodic, and clandestine detonation of nuclear weapons.

We also proceed to explore the implication of the simulated model to state-of- the-art pattern recognition methodologies