Efficient Method of Parameters Optimization of LSSVM for Large Sample Data

In order to effectively solve the problem of traditional LSSVM being computationally infeasible for large sample data, an efficient method of parameters optimization of least squares support vector machine (LSSVM) for large sample data is proposed. The regularization parameter and Gaussian kernel parameter are optimized by using the sequential optimization method based on Kriging surrogate. Three classical regression datasets, i.e., abalone, cpusmall and space_ga, are used to validate the above method. Then, a comparison is made between the prediction results obtained by using the proposed method and by directly using the LS-SVMlab toolbox. The results show that the proposed method can not only assure the prediction accuracy but also reduce the training time markedly. Finally, the proposed method is successfully applied to calculate the firing data of precision weapons.