基于决策树的快速SVM分类方法

摘要

为提高支持向量机(support vector machine,SVM)算法对大规模数据的适应能力,加快SVM算法的分类速度,提出一种基于决策树的快速SVM分类方法.该方法的重点在于构建一棵决策树,将大规模问题分解为相对简单的子问题,树中节点由线性支持向量机组成,每个节点包含一个决策超平面,分类过程取决于节点的数量.此方法在分类复杂样本时避免了使用非线性核函数.并且由于使用线性核函数,则不用进行模型选择,进一步加快了样本的分类速度.实验表明,针对大规模多特征数据的非线性分类问题,该方法比传统方法具有更高的速度.%In order to improve the large-scale data adaptability of the support vector machine (SVM) algorithm, accelerate the classification speed of the SVM algorithm, one fast SVM classification method is proposed based on the decision tree. The focus of this method is to construct a decision tree and decompose the large-scale problem into relatively simple sub-problems, the tree nodes are composed by the linear SVMs, then each node contains a decision hyperplane, the classification process depends on the number of nodes. This method avoids using the nonlinear kernel function in classification of complex samples, and by using a linear kernel function, it needs not to undertake the model selection, thus accelerating the samples classification rate. Experiments show that for the nonlinear classification problem of large-scale data with multiple features, the method has higher speed than the traditional methods.