Study of SVM decision-tree optimization algorithm based on genetic algorithm

摘要

In this paper, we present a SVM multi-classification decision-tree optimization algorithm based on genetic algorithm (GA) in order to overcome the defect of the error accumulation which is caused by the fixed tree configuration of traditional support vector machine (SVM) multi-classification decision-tree algorithms and the random positions of their decision nodes. We adopt the “classification margin” of SVM as GA adaptive function. Then, GA is used to create optimal or suboptimal decision-tree automatically, which makes the margin between two classes maximal at every decision node. Experimental results show that the error accumulation phenomenon is weakened obviously and classification quality is advanced greatly compared with the traditional algorithms.