Abstract <p>A global trend in the development of modern society is the intensive development and mass implementation of artificial intelligence technologies into practice. In this case, recognition theory is a scientific discipline whose problems concern the processes of acquiring knowledge by an intellectual system during its functioning. Pattern recognition is concerned with the development of principles, methods, and algorithms for determining whether an object belongs to one of the classes. The central place in recognition is occupied by the process of learning from precedents, the conceptual basis of which is determined by the methods of describing and dividing classes. There are three basic principles known: enumeration of class members, clustering, and commonality of properties. An approach to the implementation of the principle of common properties based on the search for combinations of features that ensure the separation of classes is proposed. An original learning algorithm is presented, aimed at automatically detecting hidden patterns in the training sample, which are then used to build a classifier. The effectiveness of the approach is confirmed by the results of a numerical experiment using model data as an example.</p>