Some Computational Considerations for Kernel-Based Support Vector Machine

Abstract

Sometimes healthcare perspectives in communications technologies require data mining, especially classification as a supervised learning. Support vector machines (SVMs) are considered as efficient supervised learning approaches for classification due to their robustness against several types of model misspecifications and outliers. Kernel-based SVMs are known to be more flexible tools for a wide range of supervised learning tasks and can efficiently handle non-linear relationship between input variables and outputs (or labels). They are more robust with respect to the aforementioned model misspecifications, and also more accurate in the sense that the root-mean-square error computed by fitting the kernel-based SVMs is considerably smaller than the one computed by fitting the standard/linear SVMs. However, the choice of kernel type and particularity kernel’s parameters could have significant impact on the classification accuracy and other supervised learning tasks required in network security, Internet of things, cybersecurity, etc. One of the findings of this study is that larger kernel parameter(s) would encourage SVMs with more localities and vice versa. This chapter provides some results on the effect of the kernel parameter on the kernel-based SVM classification. We thus first examine the effect of these parameters on the classification results using the kernel-based SVM, and then specify the optimal value of these parameters using cross-validation (CV) technique.