基于遗传算法的SVM多分类决策树优化算法研究  

Study on GA-based SVM multi-class classification decision-tree optimization algorithm

作　　者：连可[1] 陈世杰[1] 周建明[1] 龙兵[1] 王厚军[1] 

机构地区：[1]电子科技大学自动化工程学院,成都610054

出　　处：《控制与决策》2009年第1期7-12,共6页Control and Decision

基　　金：国防基础科研项目(A1420061264)

摘　　要：设计一种基于遗传算法(GA)的支持向量机(SVM)多分类决策树优化算法,以克服因传统SVM多分类决策树结构固定,单个SVM节点在树中位置随意而引起"误差积累"现象严重的缺陷.采用了SVM分类间隔作为GA适应度函数.利用GA在每一决策节点自动选择最优或近优的分类决策,最终自适应地实现了对决策树的优化.仿真实验表明,与传统方法相比,所提出的方法可使"误差积累"现象明显降低,分类质量大大提高.For the fixed tree configuration of traditional support vector machine （SVM） multi-class classification decision-tree algorithms and the random positions of their decision nodes, the error accumulate is very severity. Therefore, we present a GA-based SVM multi-class classification decision-tree optimization algorithm. We adopt the ＂margin＂ of SVM as adaption function to design GA. Then, GA is used to create optimal or suboptimal decision-tree automatically, which makes the margin between two classes maximal at every decision node. Experiment results show that the error accumulation phenomena is weaken obviously and classification quality is advanced greatly.

关 键 词：支持向量机 遗传算法 决策树 误差积累 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]