A Modified Particle Swarm Optimization Algorithm for Support Vector Machine Training

Abstract

A new modified particle swarm optimization algorithm for linear equation constrained optimization problem was put forward. And the method using this algorithm to train support vector machine was given. In the new algorithm, the particle studies not only from itself and the best one but also from other particles in the population with certain probability. This strengthened learning behavior can make the particle to search the whole solution space better. In addition, the mutation for the particle is considered in the new algorithm. The mutation operation is executed when the particle swarm becomes stagnated, which is decided by calculating the population diversity with the formula presented in this paper. For the specific constraints of support vector machine, a new method to initialize the particles in the feasible solution space was provided. The experiments on synthetic and sonar dataset classification show that our algorithm is feasible and robust for support vector machine training