UDC 004.9
SOLUTION TO THE PROBLEM OF ONE-CLASS CLASSIFICATION
IN A VARIABLE-SIZE DESIGN SPACE
USING SETS OF REGULAR EXPRESSIONS
N. A. Demidov, assistant, Department of corporate information systems, RTU MIREA, Moscow, Russia;
orcid.org/ 0009-0009-9143-6515, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
This article explores an approach to solving a mixed optimization problem in variable-size design space
using the example of tuning the parameters of one-class SVM classifier for classifying datasets of regular
expressions. The proposed approach enables switching between parameter spaces of different dimensions
during the implementation of genetic algorithm by introducing an additional gene encoding a dimensional
variable into a chromosome. The dimensional variable enables switching between optimized parameter
spaces during optimization process and distinguishes chromosome genes into active and passive ones. The
approach utilizes Gray code to encode parameters of different types in chromosome genes. Experimental
results obtained using the example of tuning the parameters of one-class SVM classifiers for datasets of regular
expressions confirm the effectiveness of the proposed approach. A genetic algorithm, in which an additional
gene encoding a dimensional variable is introduced into a chromosome, provides a simultaneous
search for the values of optimized parameters in spaces of different dimensions and allows one to obtain
high values of classification quality metrics, in particular, high values of F1-score metric on regular expressions
belonging to «novelty» class in the problem of one-class classification.
Key words: mixed optimization problem, variable-size design space, dimensional variable, genetic algorithm,
One Class SVM algorithm, Gray code, dataset, regular expression.