SVDD-based method for fast training of multi-class support vector classifier

Multi-class classifiers such as one-against-all or one-againstone methods using binary classifier are computationally expensive to solve multi-class problems due to repeated use of training data. Actually, they usually need training data for at least two classes to train each binary classifier. To solve this problem, we propose a new multi-class classification method based on SVDD (Support Vector Domain Description) that needs only one class data to describe each class. To verify the performance of the proposed method, experiments are carried out in comparison with two other methods based on binary classification: one-against-all and one-against-one. The experiment results shows that the proposed method reduces the order of training time as the size of training data increases.