Generalized non-reducible descriptors

摘要

This paper provides a generalization of non-reducible descriptors by extending the concept of distance between patterns of different classes. Generalized non-reducible descriptors are used in supervised pattern recognition problems where the feature vectors consist of Boolean variables. Generalized non-reducible descriptors are expressed as conjunctions and correspond to syndromes in medical diagnosis. Generalized non-reducible descriptors minimize the number of operations in the decision rules. A mathematical model to construct generalized non-reducible descriptors, a computational procedure, and numerical examples are discussed.