Binary tree multi-class SVM based on OVA approach and variable neighbourhood search algorithm

Abstract

In this paper, we propose and examine the performance of an approach multiclass SVM, based on binary tree and VNS algorithm. In order to solve real multiclass problems, a mapping of the original problem to several sub-problems is used to improve the performance of multiclass SVM. The proposed paradigm is composed of two steps; in the first, we start with the construction of binary tree, using a partitioning technique, where each node is a partition of two classes. In the second step, we calculate the optimal binary tree by VNS algorithm, with the aim to explore the search space and to avoid the problem of local minima, the search process of optimisation is guided by one-versus-all strategy. In the subject to improve the temporal complexity of multiclass SVM by reducing the support vectors number and decrease the recognition time. This combination leads to decrease the test time and improve the convergence of classifier.