Posterior-probability-based binary tree of support vector machine

A complete framework of binary tree of support vector machine based on posterior probability (PPBTSVM) is proposed by introducing soft labels into the enhanced binary tree of SVM (c-BTS). The soft labels are derived from the posterior probability, which in turn is determined by an empirical window-based density estimator. The structure of PPBTSVM is almost the same as that of binary tree of SVM, but the non-leaf nodes of binary tree are now PPSVM instead of hard-labeled SVM. The procedure of the proposed algorithm is given. Simulation examples show that the proposed algorithm obtains training and classification accuracy if not higher, at least comparable to those of c-BTS, while using significantly fewer binary classifiers.