Multi-class Classification on "VINE" Structure

In this paper we present a new One-Versus-All or OVA-based scheme for multi-class classification problems, aiming to reduce the training time when applying support vector machines (SVMs), particularly on large datasets. The experimental results on ten benchmark datasets show that the performance of the proposed scheme, referred to as "VINE", is comparable to that of its predecessor OVA scheme, but the former spends less training time than the latter scheme. On the problems with a large number of dimensions and instances, it is possible to combine VINE and a feature selection to obtain further speedup.