Learning performance of Gaussian kernel online SVMC based on Markov sampling

In this paper we consider the learning ability of Gaussian kernels online support vector machine for classification (SVMC) with non-i.i.d. input samples, Markov training samples. We introduce a new Gaussian kernels online SVMC algorithm with Markov selective sampling, and give the experimental researches on the generalization ability of online SVMC method with Markov selective sampling for RBF kernels and benchmark repository. The numerical studies show that the learning ability of Gaussian kernels online SVMC method with Markov selective sampling is better than that of randomly independent sampling.