Classification procedures based on the construction of class coverings
E. V. Djukova, A. S. Inyakin · 2003
—An approach to the clustering problem with integer data is described, in which special sets of feature values not contained in the feature descriptions of objects are constructed. The problem is reduced to the construction of irredundant coverings of integer matrices, which can be done by con-structing irreducible coverings of Boolean matrices. Based on a geometric interpretation of the concept of covering, a new method is proposed for constructing irreducible and minimal coverings of a Boolean matrix. The metric properties of close-to-minimal coverings of integer matrices are examined.