Tree-shaped Decision Schema for Multi-classification

Abstract

This paper presents a novel M-class classification schema. SVCMR creates a tree-shaped decision frame which contains M/2 three-separation classifiers as decision nodes. In an informed order, a series of basic classifiers are trained on the reduced dataset, and this order ensures the less number of basic models to cover decisions on all classes. SVCMR is characterized with individual parameterization of Kernel scale and penalty coefficient. Experiments on real experiments demonstrate the fine performance of SVCMR.