Deterministic Recursion to Target Class Classification

• Published in: Pattern recognition and image analysis Vol. 33; no. 3; pp. 584 - 598
• Main Authors: Aslanyan, Levon, Gishyan, Karen, Sahakyan, Hasmik
• Format: Journal Article
• Language: English
• Published: Moscow Pleiades Publishing 01.09.2023; Springer Nature B.V
• Subjects: Algorithms; Application Problems; Classification; Cluster analysis; Computer Science; Graph theory; Image Processing and Computer Vision; Machine learning; Mathematical analysis; Mathematical models; Optimization; Pattern Recognition; Supervised learning; treatment regime; target class; classification; recursion
• ISSN: 1054-6618; 1555-6212
• DOI: 10.1134/S1054661823030033

Formal mathematical description and analysis of the mathematical model of the problem from the field of medicine about adaptation of dynamic treatment regimes in terms of classification algorithms, and class transition logic, is the main goal of this work. We call these models and procedures classification to the target class, because the global goal of the problem is to assign objects/patients to the normal/healthy class. The usual classification procedure that appears in the first step of the considered procedures is supplemented by the mechanism of bounded transition between classes (boundedness is a property of the considered domain), and this process consisting of (object-action) pairs is repeated recursively, seeking to assign all objects to the same predetermined target class. The correspondences and differences of the obtained model with the main classes and approaches of machine learning—supervised learning, cluster analysis, and reinforcement learning—are analyzed. The class of emergent problems such as model validation and optimization is considered. In particular, a logical analysis of the system of class transitions is given and an augmented graph-theoretic solution for the deterministic case of target class classification problem transitions is obtained.