One-Class Support Vector Machine Calibration Using Particle Swarm Optimisation

Abstract

Abstract. Population-based search methods such as evolutionary algorithms, shuffled complex algorithms, simulated annealing and ant colony search are increasingly used as automatic calibration methods for a wide range of numerical models. This paper proposes the use of particle swarm optimisation to calibrate the parameters a one-class support vector machine. This approach is developed and tested in the calibration of a one-class SVM, applied to several data sets. The results indicate that the proposed method is able to match or surpass the performance of a one-class SVM with parameters optimized using a standard grid search method, with much lower CPU time required.