基于核聚类的SVM多类分类方法

对支持向量机的多类分类问题进行研究，提出了一种基于核聚类的多类分类方法。利用核聚类方法将原始样本特征映射到高维特征进行聚类分组，对每一组使用一个支持向量机二值分类器进行分类，并用这些二值分类器组成决策树的节点，构成了一个决策分类树。给出决策树的生成算法，提出了利用交叠系数来控制交叠，从而克服错分积累，提高分类准确率。实验结果表明，采用该方法，手写体汉字识别速度和正确率都达到了实用的要求。

SVM multi-class classification based on kernel clustering algorithm

The Method of multi-class classification based on Support Vector Machines(SVM) was researched,and a multi-class classified method based on kernel clustering algorithm was proposed.The original features were mapped to higher dimension applying kernel clustering;each group was classified by a SVM classifier.These binary classifiers were seen as the node of decision tree to construct a decision classifying tree.The algorithm of creating decision tree was given in this paper.The method of controlling overlap by overlap coefficient was proposed and it can overcome accumulating misclassification.Experiments show that with the approach in this paper,the speed and accuracy rate of handwritten Chinese character recognition can satisfy the functional need.