Development of the SVM classifier by means of the hybrid versions of the particle swarm optimization algorithm based on the grid search
Liliya Demidova, Irina Klyueva
COMPUTER MODELLING & NEW TECHNOLOGIES 2017 21(1) 56-63
Ryazan State Radio Engineering University Gagarin Str., 59/1, Ryazan, Russian Federation
In this article the approaches to the problem solving of searching of the parameters of the SVM classifier based on the hybridization of the particle swarm optimization algorithm (PSO algorithm) and the grid search algorithms with the aim of providing of high quality classification decisions have been considered. The paper presents two hybrid versions of the basic PSO algorithm, involving the use of the classical Grid Search (GS) algorithm and Design of Experiment (DOE) algorithm correspondingly. It is proposed to use the canonical PSO algorithm as the basic algorithm. The results of experimental studies confirm the application efficiency of the hybrid versions of the basic PSO algorithm with the aim of reducing of the time expenditures for searching the optimum parameters of the SVM classifier while maintaining of high quality of its classification decisions.