Abstract:

In practical applications, the performance of Support Vector Machine (SVM) depends on the selection of parameters. SVM parameter selection problems are studied and analyzed, and a parameter optimization method of SVM based on uniform design is proposed. Our method is compared with the parameter optimization methods of SVM based on grid search, particle swarm optimization, and genetic algorithms. Multiple classification data sets with different sizes are used for testing, so as to compare the classification accuracy and runtime of the four methods. Simulation results show that all the four methods can find the optimal parameters, which make the classification accuracy of SVM approach or exceed the theoretical accuracy of categorical datasets, and our proposed method has the characteristic of finding parameters in short time.

Key words: support vector machine;parameter optimization method;uniform design;grid search;particle swarm optimization;genetic algorithm