Novel Multiclass SVM-Based Binary Decision Tree Classifier

This paper proposes a novel algorithm for constructing multiclass SVM-based binary decision tree classifiers. The basic strategy of the proposed algorithm is to set the target values for the training patterns such that linear separability is always achieved and thus a linear SVM can be constructed at each non-leaf node. It is argued that replacing complex, nonlinear SVMs by a larger number of linear SVMs remarkably reduces training and classification times as well as classifier size without compromising classification performance. This is experimentally demonstrated through a comparative analysis involving the most efficient existing multiclass SVM classifiers, namely the one-against-rest and the one-against-one.