Parameters optimization of Support Vector Machine based on Simulated Annealing and Genetic Algorithm

The generalization error of support vector machine usually depends on its kernel parameters, but there is no analytic method to choose kernel parameters for SVM. In order to choose the kernel parameters for SVM, the simulated annealing algorithm and genetic algorithm are combined, which is called simulated annealing genetic algorithm (SA-GA), to choose the SVM kernel parameters. SA-GA makes use of encoding method, reproduction, crossover and mutation in the SA when generate new solution. In this way, the characteristic of SA that can accept a worse solution in a certain extent of probability can solve premature convergence of GA, and the heuristic search method of GA can make SA robust to the parameters of cooling schedule. So the combined algorithm has better performance than SA or GA, and it can get a better solution for optimization problem. At last, SA-GA has been used to choosing the kernel parameters of SVM. The results of simulation show that the performance of the method that proposed in this paper was more efficient than SA and GA for choosing kernel parameters of SVM.