用于多类分类的层次式支持向量机

Hierarchical Support Vector Machine for Multi-class Classification

针对支持向量机存在的训练数据量大导致的训练时间过长和训练数据不平衡导致的分类结果会向训练数据多的类倾斜等问题，提出了适合于多类分类的层次式支持向量机。在训练过程中，首先折衷考虑各类之间的距离和各类的训练数据长度，据此将训练样本分为距离较远且其长度基本平衡的2类，然后逐层进行训练，最终形成二叉树分类结构。仿真实验证明，该方法能够有效地缩短训练和分类时间，且对多类分类中的数据不平衡问题有一定的效果。

In the process of training Support Vector Machine Classifier, the training time will be much longer when training data are large, and the accuracy of each class will be inclining towards the class that has more training data when training data are uneven. The layers SVM are presented in this paper. In the process of training, the distances between classes and lengths of each class are considered, and training data are split into two classes whose the distance is longer and length is in balance. The simulation validates the training time would be shorter and the problem of uneven training data could be alleviated.