Precision of multi-class classification methods for support vector machines

A single support vector machines can only deal with binary-class classification. In the face of multi-class classification task, we usually combine multi-support vector machines together, strategies include one-against-one, one-against-the-rest and Binary Tree, etc. Previously precisions of these classification methods are drawn from experiments. In this paper we propose precision formulae of one-against-the-rest and Binary Tree classification methods, experiments show our conclusions are convincing.