基于熵变的DAG-SVMs的组合策略

Composing strategy of entropy-based DAG-SVMs

为了降低传统的有向无环图支持向量机(DAG-SVMs)多类分类方法在模型构建过程中节点选择的随机性,提高最终分类结果的准确率,提出了一种基于熵变的有向无环图支持向量机(E-DAG-SVMs)的组合策略。该策略通过计算各支持向量机在分割样本集时引发的熵变,依据信息增益最大化原则来决定节点的选择,进而构建多类分类模型。实验结果表明了该策略的有效性。

To reduce the node choice randomicity of traditional directed acyclic graph support vector machine（DAG-SVMs） multiclassification model,improve the final classification accuracy,a composing strategy of the entropy-based DAG-SVMs（E-DAG-SVMs） is proposed.This strategy calculates the change of entropy,when each support vector machines splitting the sample set,determines the choice of node based on the principle of maximization of information gain,constructs the multi-classification model.By the comparison experiment to the other methods mentioned,the validity of this strategy is demonstrated.