Support vector machine with parameter optimization by a novel hybrid method and its application to fault diagnosis

• A new hybrid method is used to optimize the parameters of SVM.
• ICDF measure is used to determine a small yet effective search interval for kernel parameter.
• An almost parameter-free evolutionary algorithm, BBDE, is used to search the optimal parameters of SVM.
• The proposed method is tested on several benchmark datasets as well as fault diagnosis for rolling element bearings.
• Experimental results show the outstanding performance of the proposed method.

The performance of support vector machine (SVM) heavily depends on its parameters. The parameter optimization for SVM is still an ongoing research issue. The current parameter optimization methods either are easy to fall into local optimal solution, or are time consuming. Moreover, some optimization methods depend also on the choice of parameters for them, provoking thus a vicious circle. In view of this, a new hybrid method is proposed to optimize the parameters of SVM in this paper. It uses the inter-cluster distance in the feature space (ICDF) to determine a small and effective search interval from a larger kernel parameter search space, while a hybrid of the barebones particle swarm optimization and differential evolution (BBDE) is used to search the optimal parameter combination in the new search space. The ICDF shows the degree the classes are separated. The BBDE is a new, almost parameter-free optimization algorithm. Some benchmark datasets are used to evaluate the proposed algorithm. Furthermore, the proposed method is used to diagnose the faults of rolling element bearings. Experiments and engineering application show that the proposed method outperforms other methods both mentioned in this paper and published in other literature.