Research On Modeling Method Of Support Vector Machine

Support vector machine (SVM), the kernel content of the Statistical Learning Theory, is a new and outstanding machine learning method, which was put forward in the twentieth century. It is a valid machine-learning tool in dealing with small samples. SVM overcomes some shortcomings of neural network, such as slow convergence, unstable solution and poor generalization. So it has been widely applied to many areas, such as pattern recognition, signal processing, automation and communication. So it is very important to study the theory and its application for researching and producing purpose.SVM for regression is an important researching field. In this paper, several regression algorithms based on SVM are proposed. The performance and applications of the algorithms are studied in six chapters. The research is carried out in the following aspects:1.An overview on machine learning, statistical learning theory and structural risk minimization is provided, followed by a summary on the development and applications of SVM.2. Based on the analysis of the basic theory of SVM, the improved algorithms of SVM are summed up and compared with traditional one.3.A novel kernel function is put forward in solving nonlinear modeling problem, which overcomes the defect of Gauss kernel function when the support vectors are too close to one another. Simulation results show good performance in forecasting error.4.A modified SVM algorithm for regression inspired from the optimization of Fisher's discriminant ratio is presented, which can reduce the wrongly scattered rate by using between class scatter matrixes. The advantage is proved by theoretic and simulation study.5.A framework for incremental algorithm of SVM for regression is presented, which has a better performance than traditional one in training speed when the forecasting error is achieved.