Parameters selection for SVM using simulated annealing combinatorial algorithms
Nie Jingxu · Computer Engineering and Applications Journal · 2010
Considering the deficiencies of traditional ways,one sort of combinatorial simulated annealing algorithms is introduced to establish a new method for the SVM parameters selection.In this method,appropriate objective function is set to guarantee SVM'maximum generalization ability.Meanwhile,by referring to the cross-validation principle,are the samples in the training set utilized to select SVM models,and the samples in the testing set to search optimal parameters.Finally,a comparative analysis is conducted upon the data of simulation experiments between the proposed approach and those based on genetic algorithms and refined mesh algorithms.The results show that this method is endowed with a better global search performance and a higher convergence rate,which make it an effective way to determine optimal SVM parameters,and therefore enjoys a strong practical value.