SVM Parameter Optimization And Its Application In The Classification

Support vector machine is a machine learning algorithm based on the statistical learning theory proposed by Vapnik in the mid-1990s, and it has been used in many fields. In SVM applications, the kernel function and its parameters have a great impact on classification results. Therefore, parameters optimization of support vector machine has become a concerned research topic in SVM research.In the begining, the parameters in SVM are adjusted manually; in recent years, some intelligent algorithms are used to optimize the parameters in SVM, such as genetic algorithms, particle swarm optimization and grid search algorithm and so on.The main tasks of this paper are as follows:Firstly, parameters optimization of support vector machines. Gentic algorithm (GA), particle swarm optimization algorithm (PSO) and grid search algorithm (GSA) are used to optimize the parameters in SVM. The parameters are the penalty parameter C and the parameter σ in gaussian kernel function. Secondly, classification based on SVM with optimized parameters. The parameters in SVM are optimized using GA, PSO and GSA. SVMs with optimized parameters are applied on UCI datasets. The results show the advantages and disadvantages of the three optimization algorithms. When the dataset is small (its sample number is small), GSA performs best. When the dataset is large (its sample number is large), GA and PSO are better. Compared with PSO, GA has a higher classification accuracy, although it is more time-consumed.