Support Vector Machines with Binary Tree Architecture for Multi-Class Classification
SungMoon Cheong, S. H. Oh, Soo-Young Lee · International Conference on Neural Information Processing · 2004
For multi-class classification with Support Vector Machines (SVMs) a binary decision tree architecture is proposed for computational efficiency. The proposed SVM- based binary tree takes advantage of both the efficient computation of the tree architecture and the high classification accuracy of SVMs. A modified Self-Organizing Map (SOM), K- SOM (Kernel-based SOM), is introduced to convert the multi-class problems into binary trees, in which the binary decisions are made by SVMs. For consistency between the SOM and SVM the K-SOM utilizes distance measures at the kernel space, not at the input space. Also, by allowing overlaps in the binary decision tree, it overcomes the performance degradation of the tree structure, and shows classification accuracy comparable to those of the popular multi-class SVM approaches with one-to-one and one-to-the others. Keywords—Support Vector Machine, multi-class classification, Self-Organizing Map, binary decision tree