| Support vector machine (SVM) distinguishes itself among other methods in research of epileptic EGG for its excellent performance.SVM is a new machine learning method. It is a collection of several criterions and technologies in the area of machine learning method. It integrates the techniques of maximal interval hyperplane, Mercer kernel, convex quadratic programming, sparse solution, slack variable and so on. The least square support vector machine (LS-SVM) is an improved method based on the SVM. Combining least square method, wavelet kernel function and support vector machine, the performance of LS-SVM on rate, precision and generalization ability are better off compared with the standard SVM. However, there still exist some problems when determining the parameters for the model. The number of parameters in LS-SVM is more than the one of SVM, while they have enormous effect on the performance. Up to now, there is no theory for parameters determining.The major contents of this thesis are as follows:1. Firstly the theory, mathematics model and structure of the SVM are studied, then the principle, characteristic and operation method of genetic algorithm (GA) are researched. At last the combination of SVM and GA is proposed.2. A genetic algorithm is developed to optimize the parameters of the LS-WSVM. Then the genetic algorithm LS-WSVM (GA-LS-WSVM) is constructed. Two experiments are performed to validate the effectiveness of the algorithm.3. Finally, a method of EEG signal classification between healthy people and epileptics based on GA-LS-WSVM is proposed. The data analysis shows that the performance is excellence.This paper proposes the genetic least square wavelet support vector machine to overcome the parameter selection problem. Simulation results show out that it has better performance than genetic least square wavelet support vector machine and least square support vector machine. Therefore, it can be better applied in engineering practice. |
PDF Full Text Request