Research Of The Optimization Of SVM Parameter Based On PFO

Support Vector Machine (SVM) is a high-Performance learning machine which has been developed on the basis of the statistical learning theory. SVM is used to find the best compromise between the complexity of the model and the learning ability according to the information of limited samples to obtain the best generalization ability. There is a problem in SVM of which the performance depends on the setting of parameters, including penalty factors and kernel parameters, but there is no suitable theory that could be the guide to find adapted parameters for concrete sample data. The selection of parameters is an important issue of the research in Support Vector Machine, whose essence is a procedure of optimization.Particle Swarm Optimization (PSO) is a new branch of Swarm Intelligence which was originated by the research in prey behavior of birds swarm. It was first introduced in1995by Kennedy and Eberhart. The algorithm is used to find the best solution through the cooperation of individuals. It is simple and easy to be implemented, yet has problems which are premature convergence and tending to get stuck in local minima. As for these problems, Particle Filter Optimization (PFO) algorithm of Particle Swarm Optimization combining with Particle Filter (PF) is presented. Experiments show that PFO is better than PSO in both convergence rate and final suitable value.I first analyzed the effect of penalty factors and kernel parameters in the SVM model, introduced the existing optimization methods of parameters. Then, according to the Particle Swarm Optimization (PSO) method in SVM parameters optimization, I put forward a new algorithm based on improved Particle Filter Optimization (PFO) method in SVM parameters optimization. Finally, I investigated electrocardiogram classification based on Support Vector Machine algorithm on the MIT-BIH database. The reasonable penalties and kernel parameters are chosen from the two optimization algorithms. I use samples to train the SVM classifier to obtain the results. After comparing, We made a conclusion that PFO-SVM is better than SVM and PSO-SVM in speed, global search abilities and accuracy of classification.