Supervised pattern recognition by parallel feature partitioning

In the present paper the supervised pattern recognition problem is considered. For solving the problem a mathematical model based on parallel feature partitioning is proposed. The solution is obtained by partitioning the feature space to a minimal number of nonintersecting regions. This is achieved by solving an integer-valued optimization problem, which leads to the construction of minimal covering. Since the classes do not intersect it follows that the solution of the formulated problem exists. Computational complexity of the model and computational procedures are discussed. Geometrical interpretation of the solution is given. (C) 2003 Pattern Recognition Society. Published by Elsevier Ltd. All rights reserved.