Metaheuristic techniques for Support Vector Machine model selection

Abstract

The classification accuracy of a Support Vector Machine is dependent upon the specification of model parameters. The problem of finding these parameters, called the model selection problem, can be very computationally intensive, and is exacerbated by the fact that once selected, these model parameters do not carry across from one dataset to another. This paper describes implementations of both Ant Colony Optimization and Particle Swarm Optimization techniques to the SVM model selection problem. The results of these implementations on some common datasets are compared to each other and to the results of other SVM model selection techniques.