Support vector machines are based on the theoretical learning theory developed by Vapnik [12], [17, pp. 92-129], [48], who defies the conventional belief that the optimal classification system can be developed using the optimally reduced features. In support vector machines, an n-class problem is converted into n two-class problems in which one class is separated from the remaining classes. For each two-class problem, the original input space is mapped into the high dimensional dot product space called feature space and in the feature space, the optimal hyperplane that maximizes the generalization ability from the stand point of the VC dimension is determined.