Learning performance of multi-class support vector machines based on Markov sampling

SVM was originally introduced for classification problem with two class under the condition that the input samples are drawn independent and identically distributed (i.i.d.) from a given data. SVM had been considered to research the multi-class classification problem by solving a series of b classification problems with two class such as the “one-against-one” (OAO) algorithm and the “one-against-all” (OAA) algorithm. In this text, we research the multi-class support vector machine for classification with based on Markov selective sampling and the OAO method. We first introduce a new OAO multi-class SVMC algorithm based on Markov sampling and give the experimental researchs on the learning ability of OAO multi-class SVMC with Markov selective sampling based on real-world data sets. These experimental researchs indicate that the learning performance of the OAO multi-class SVMC with Markov selective sampling is better than that of random sampling.