Abstract

We mainly study the voting model. The article considers recognition problems with disjoint classes. These problems are, in a particular case, a discrete analogue of the problem of finding optimal solutions. Not only the problems of synthesizing the best solutions, but also other important classes of applied problems are reduced to recognition problems. In real calculations, there is no need to remember all the parameters ${{P}_{rv}}\cdot {{\varepsilon }_{rv}}$ that determine the proximity function for the recognizable object and the sets ${{K}_{j}}$. It is enough to limit ourselves to only a small part. The values ​​of the parameters ${{\varepsilon }_{ik}}{{P}_{ik}}$ in problems with disjoint classes are determined independently for each class.

In this paper, we describe methods that allow you to select the parameters ${{\varepsilon }_{ik}}{{P}_{ik}}$ depending on the values ​​of the $k$-th feature ${{a}_{ik}}$ on the objects of the original information. An algorithm for selecting the parameters ${{P}_{ik}}$ that determine the proximity function for the recognizable object and the classes ${{K}_{j}}$ has been developed. and the choice of parameters ${{\varepsilon }_{ik}}$, defining the proximity function in recognition problems with non-overlapping classes. A method for constructing a support set for a recognition algorithm in problems of classifying objects with disjoint classes has been proposed. A numerical method for finding optimal values ​​of the parameters ${{\varepsilon }_{{{i}_{k}}}}{{P}_{{{i}_{k}}}}$ defining the proximity function in recognition problems with non-overlapping classes based on solving systems of Boolean equations and searching for irreducible table coverage has been developed.

Statistics