Optimizing parameters of support vector machines using team-search-based particle swarm optimization

It is greatly important to select the parameters for support vector machines (SVM), which is usually determined by cross-validation. However, the cross-validation is very time-consuming and complicated to create good parameters for SVM. The parameter tuning issue can be solved in the optimization framework. The paper aims to discuss these issues.

In this paper, the authors propose a novel variant of particle swarm optimization (PSO) for the selection of parameters in SVM. The proposed algorithm is denoted as PSO-TS (PSO algorithm with team-search strategy), which is with team-based local search strategy and dynamic inertia factor. The ultimate design purpose of the strategy is to realize that the algorithm can be suitable for different problems with good balance between exploration and exploitation and efficiently control the inertia of the flight. In PSO-TS, the particles accomplish the assigned tasks according to different topology and detailedly search the achieved and potential regions. The authors also theoretically analyze the behavior of PSO-TS and demonstrate they can share the different information from their neighbors to maintain diversity for efficient search.

The validation of PSO-TS is conducted over a widely used benchmark functions and applied to tuning the parameters of SVM. The experimental results demonstrate that the proposed algorithm can tune the parameters of SVM efficiently.

The developed method is original.