A new GA-based decision search for DAG-SVM

For an N-class problem, the decision directed acyclic support vector machines (DAG-SVM) construct N(N-1)/2 classifiers, one for each pair of classes. But the generalization performance of the original DAG-SVM depends a lot on the nodes sequence of the directed acyclic graph. To get a good generalization performance, genetic algorithm is used to permute searching nodes in a DAG. For any test sample, the decision process is intelligent based on the searching sequence obtaining from genetic algorithm. Experiments show the efficiency and feasibility of the new approach.