Research Of Parameter Selection For Support Vector Machine

Support Vector Machine (SVM) is a machine learning method which is widely used at present. This method when the sample is not enough still can show the good effect, it now become the research field of machine learning. Numerous studies show that support vector machine parameters selection has a great influence for the classification effect. Support vector classification machine need to optimize the two parameters that are the penalty parameter C、kernel parameter σ.But now there is no theory to guide parameter selection. Commonly used methods include experiment method, grid method, gradient descent method, intelligent optimization algorithm, etc. Most of the methods are to choose two parameters as a whole. However, these methods operating process is often cumbersome and the effect is not very ideal. In this paper, the penalty parameter and kernel parameter respectively use different optimization rules. It then simplifies the operation process of parameter choice and improves the classification performance of SVM model. The researches of the thesis are follows:(1) By introducing the Support Vector Machine theory, then it introduces the basic concept of the support vector machine including the optimal separating hyperplane, linearly separable support vector machine, linear inseparable support vector machine, nonlinear support vector machine.(2) It respectively introduces the nuclear and penalty parameters on the influence of the support vector machine and introduces several common optimization parameters of the comparison method:genetic algorithms, particle swarm optimization, grid search algorithm. And then put forward the parameter selection method of this paper which is to study the matrix similarity and the deficiencies. Improved problem is become a mixed containing nuclear parameter expression of maximum. And then use genetic algorithm to solve it to get kernel parameter. Then it gets the optimal kernel parameter into the support vector machine model to take advantage of hybrid intelligence algorithm to get penalty parameter.(3) Experimental simulation results show that the new method proposed in this paper which choose the parameters are applied to support vector classification machine model, class performance of the classifier has a lot of ascension. At the same time, the method to establish the time of the SVM model is relatively shorter.