基于GA和KNN的SVM决策树分类方法研究

Research of SVM Decision-treen Classification Based on GA and KNN

文章提出了一种基于遗传算法和K近邻的SVM决策树方法,并将其应用于解决SVM多分类问题。算法以基于类分布的类间分离性测度为准则,利用遗传算法对传统的SVM决策树进行优化,生成最优(较优)决策树。在分类阶段,对容易分的节点利用SVM进行分类,而对可分离性差的节点采用SVM和K近邻相结合的分类方法,最终实现多类别分类。实验结果表明,与传统的分类方法相比,该算法的实验效果较好,是一种有效的分类方法。

In this paper,a SVM decision-tree algorithm was presented based on GA and KNN.First,GA is used to create optimal or near-optimal decision-tree,which defines a novel separability measure.Then in the class phase,standard SVM is used to make binary classification for the divisible nodes,and SVM combined with KNN are used to classify the fallible nodes.Finally,the multi-classification is achieved by the SVM decision-tree.Experimental results show that the proposed method could effectively improve the classification precision in comparison to traditional classification methods.