A KKT Condition Based DDAGSVM Classifier

Decision directed acyclic graph support vector machine (DDAGSVM) has been proposed to extend SVM from binary classification problems to multi-class classifications. But the generalization ability is subject to the structure of DDAG. To improve the classification accuracy, a novel separability measure is defined based on Karush-Kuhn-Tucher (KKT) condition, and an improved DDAGSVM has been given. The experimental results show that this algorithm has higher generalization ability.