【摘要】 针对一般的SVM方法不能有效地处理不平衡样本数据及现有的偏二叉树结构SVM分类器速度慢的这两个问题,提出了一种基于球结构的完全二叉树SVM多分类算法。该算法利用球结构的SVM考虑了每个类的分布情况,能有效地处理不平衡样本数据;构建完全二叉树结构,使得同层节点所代表的SVM分类器可以并行工作,能提高其训练和分类速度,分类速度相当于折半查找。实例验证两者结合后的算法可实现准确且高效的多类分类。更多还原

【Abstract】 Aiming at the two problems which are generic SVM algorithm can not effectively dispose training sets with uneven class sizes and SVM classifiers of existing partial-binary trees have lower velocity,proposed a SVM multi-class classification algorithm based on full-binary tree of sphere-structured.Sphere-structured SVM of this algorithm can dispose unbalanced samples data because it considered the distribution of each class.To build the structure of full-binary tree,it made SVM classifiers which were denoted by the same hiberarchy nodes work at one time,which speeded up training and classification of SVM classifiers and the speed of classification was the same as bisearch.The results of example show that the proposed method can implement more exact and effective classification.更多还原