Target Class Classification Recursion Preliminaries

In traditional machine learning, geometrically "compact" sets of classes are given by representatives with the aim of correctly classifying new objects into classes. We consider an application where one class is singled out, and the goal is to classify objects into this target class by a s...

Full description

Saved in:
Bibliographic Details
Published inBaltic Journal of Modern Computing Vol. 11; no. 3; pp. 398 - 410
Main Authors Aslanyan, Levon, Gishyan, Karen, Sahakyan, Hasmik
Format Journal Article
LanguageEnglish
Published Riga University of Latvia 2023
Subjects
Algorithms
Classification
Machine learning
Precision medicine
Online AccessGet full text

Cover

More Information
Summary:In traditional machine learning, geometrically "compact" sets of classes are given by representatives with the aim of correctly classifying new objects into classes. We consider an application where one class is singled out, and the goal is to classify objects into this target class by a stepwise procedure. This problem appears in the precision medicine area under the name of "dynamic treatment regime". A predefined class-action is applied that transforms the object to the same or some other class. The mapping of class-actions to the classes forms the policy, and the chain of policy-based transformations (under some restrictions) must converge to the target class. With graph theoretical tools, we present and evaluate the policy form, and show that the graph is partitioned into a couple of structural components, which helps to find out the possible policy defects to be corrected by subject domain specialists for better results.
ISSN:2255-8950
2255-8942
2255-8950
DOI:10.22364/bjmc.2023.11.3.03