A grid-based ACO algorithm for parameters optimization in support vector machines

Abstract

The parameters optimization of the penalty constant C and the bandwidth of the radial basis function (RBF) kernel sigma is an important step in establishing an efficient and high-performance support vector machines (SVMs) model. Aiming at optimizing the parameters of SVMs, this paper presents a grid-based ant colony optimization (ACO) algorithm to choose parameters C and sigma automatically for SVMs instead of selecting parameters randomly by humanpsilas experience, so that the generalization error can be reduced and the generalization performance can be improved simultaneously. Some experimental results confirm the feasibility and efficiency of the approach.