A Study Of The New Improved SVM On The Basis Of A New Composite Kernel Function And Soft Margin

This paper summarized the support vector machine (SVM) theory. Support vector machine is based on Statistical Learning Theory (SLT). Vapnik is the person who proposed the SVM Learning method .SVM's the optimal separating hyperplane is proposed from linear divisible case. And its rapid development and the wide application are due to the introduction of the concept of soft margin.After a further study of the support vector machine theory and research status at home and abroad, the working principle and structure of SVM kernel function and SVM soft margin strategy get a deeper study.However, the traditional soft margin SVM gives the same misclassification costs for the various sample data, thus the processing results of the real data are not satisfactory. And the traditional single kernel function SVM owns bad balanced performances between the learning ability and the generalization ability.In this paper, the traditional SVM algorithm is improved by proposing two improvements. First of all, the traditional SVM soft margin algorithm is improved by paying different costs for different misclassification. Secondly, a new composite kernel function is proposed based on the original kernel function, and it owns the more balanced learning ability and generalization ability by adjusting the new kernel's parameters.A new improved SVM pattern is established on the base of a new composite kernel function and soft margin. The performance of the new improved SVM is verified through experiments. Then the improved SVM are applied to practical problems and the classification results are analyzed comparatively.The last chapter is a summary of the paper and the future outlook of the developments in the field.