Logical classification based on finding regular representative elementary classifiers

An approach to the supervised classification problem based on the apparatus of discrete mathematics (logical methods of data analysis) is considered. The possibility of time costs reducing at the stage of correct logical classifier training is investigated. New models of classifiers are proposed. These models are based on finding frequently occurring fragments of a special type in the descriptions of precedents — regular elementary classifiers. Descriptions of classifier models are given using the concepts of logical functions theory. To construct sought fragments, the authors have developed and implemented an original algorithm. The effectiveness of proposed classifier models has been experimentally substantiated and confirmed by theoretical estimates of their training complexity. An upper asymptotic estimate of the typical number of regular elementary classifiers is obtained.