Optimization of Multi-class SVM Binary Tree based on Binary PSO

摘要

Multi-class SVM classification binary tree has the problem of "error accumulation",the classification error caused by the upper node of the binary tree can't be corrected by the following nodes,this problem reduce the classification accuracy. In order to overcome the "error accumulation " produced by the binary tree of SVM multi-class classification,the binary PSO algorithm (BPSO) was introduced to optimize every node of multi-class SVM classification binary tree by designing particles encoding and fitness function of binary PSO algorithm and building the binary tree during the optimization process. Experiments show that the multi-class SVM classification binary tree optimized by the binary PSO achieves better multi-class classification accuracy,and the binary BPSO reduces the time needed to optimize binary tree.