Block Factor Based Multi-population PSO and Its Application in SVM

Abstract

In view of the low information transmission efficiency and tendency to fall into local optimal solution of the problems of traditional particle swarm optimization (PSO), an improved bidirectional information transmission block factor based multi-population PSO(BFM-PSO) is proposed. We use the improved BFM-PSO algorithm to optimize SVM penalty factor C and the radius s of the radial basis function. Tests In a number of public data sets show that SVM optimized by the improved BFM-PSO can effectively improve the classification accuracy.