Correct supervised classification: JSM-method over product of partial orders

The authors consider a logical approach to the supervised classification problem. A difference and a connection between two well-known directions of logical classification are noted: the direction represented by Correct Voting Procedures (CVP) and the direction based on the ideas of the JSM-method. The issues of improving the classifiers of the second direction are considered on the basis of using a less strict decisive rule and generalizing a scheme of work in the case when the featured descriptions of the objects under study are the elements of the Cartesian product of finite partially ordered sets. The developed new models of the JSM-classifiers use the ideas proposed earlier when creating similar algorithms for the CVP direction. The results of an experimental study on real-world tasks using a special linear ordering of feature values are presented.