最大间隔最小体积球形支持向量机

Maximal-margin minimal-volume hypersphere support vector machine

结合支持向量机(SVM)类间最大分类间隔和支持向量数据描述(SVDD)类内最小描述体积思想,提出一种新的学习机器模型——最大间隔最小体积球形支持向量机(MMHSVM).模型建立两个大小不一的同心超球,将正负类样本分别映射到小超球内和大超球外,模型目标函数最大化两超球间隔,实现正负类类间间隔的最大化和各类类内体积的最小化,提高了模型的分类能力.理论分析和实验结果表明该算法是有效的.

Inspired with the ideas of support vector machine （SVM） between-class maximal classfication margin and support vector data description （SVDD） within-class minimal description volume,a novel learning machine model,maximal-margin minimal-volume hypersphere SVM（MMHSVM）,is proposed in this paper.Two different concentric hyperspheres are builted in the model,positive samples are packed in small hypersphere and negative samples are excluded outside large hypersphere.The between-class margin is maximized by model objective function,which realizes the maximization of between-class margin and the minimization within-class volume,and the model classification performance is improved.Theoretical analysis and experimental results show the effectiveness of the proposed method.