CASPIAN JOURNAL
MANAGEMENT AND HIGH TECHNOLOGIES
Use of the modified particle swarm optimization algorithm in the SVM-classifier development problem
 |
Read | Demidova L.A., Sokolova Yu.S. Use of the modified particle swarm optimization algorithm in the SVM-classifier development problem // Caspian journal : management and high technologies. — 2016. — №1. — pp. 26-38. |
Demidova L.A. - D.Sc. (Engineering), Professor, Ryazan State Radio Engineering University, 59 / 1 Gagarin St., Ryazan, 390005, Russian Federation, liliya.demidova@rambler.ru
Sokolova Yu.S. - Senior Lecturer, Ryazan State Radio Engineering University, 59 / 1 Gagarin St., Ryazan, 390005, Russian Federation, JuliaSokolova62@yandex.ru
For the problem of development of the SVM classifier providing high quality of data classification the task of development of the modified particle swarm optimization algorithm, which carries out the simultaneous search of the kernel function type, values of the kernel function parameters, value of the regularization parameter, and also the set of the support vectors corresponding to them, is considered. In a basis of the offered particle swarm optimization algorithm the idea about В«regenerationВ» of particles is put. Realization of the algorithm assumes that some particles with the worst values of the accuracy indicator of data classification (AIDC) can change type of their kernel function to the type which corresponds to the particle with the best value of the AIDC. The results of the comparative analysis of the traditional and modified particle swarm optimization algorithms received during the experimental studies are given. These results confirm the expediency of use of the modified particle swarm optimization algorithm for the purpose of reduction of time expenditure for development of the required SVM classifier.
Key words: SVM-алгоритм, разделяющая гиперплоскость, опорные векторы, классификация, оптимизация, параметры функции ядра, параметр регуляризации, алгоритм роя частиц, SVM-algorithm, separating hyperplane, support vectors, classification, optimization, parameters of