Improved RBF-SVM based on genetic algorithm and its applications

The character of RBF kernel in support vector machine was discussed, and a conclusion was drawn that the generalization ability of support vector machine could be improved by giving larger kernel parameters to those features useless for the classification problem to lower their influence on kernel function. On the basis of this conclusion, an improved multi-kernel-parameter support vector machine with RBF kernel based on genetic algorithm was proposed, where genetic algorithm was applied to find optimum kernel parameters by minimizing validation error. Experiment results of rolling bearing fault diagnosis show that the improved multi-kernel-parameter support vector machine possesses better generalization ability than conventional support vector machine does, and the kernel parameters directly reflect the classification ability of corresponding features.