Abstract
An approach to the correct supervised classification problem based on the application of logical data analysis methods is considered. The description of the operation scheme of logical classifier models aimed at constructing special fragments of precedent descriptions, called correct elementary classifiers, is provided. More complex models, namely models of logical correctors, are based on the synthesis of families of correct sets of elementary classifiers. Unlike classical models, logical correctors show good results in the case of multivalued features, i.e. features with a large number of different values. The article examines issues related to reducing time costs and improving the quality of classification of logical correctors. New deterministic and stochastic variants of such models are proposed, designed to work with partially ordered data. The results of experiments on model and real data are presented.