One-Class SVM, which is one of the important kernel methods, does not take information of data distribution into full consideration, so the improvement of its generalization performance is effected. Aiming at this issue, in this paper the authors first redefine the distance from the origin to the hyperplane, and then present a novel algorithm called One-Class SVM based on Data Distribution (DD One-Class SVM ). At last, the optimization of the DD One-Class SVM algorithm is formulated, and how to deal with the case including the nonlinear one, where the scatter matrix is singular, is discussed in detail. In contrast to the traditional One-Class SVM , the proposed method shows better generalization ability.%单类支撑向量机(One-Class SVM)是一种重要的核方法,但是其缺少对数据分布信息的考虑,因而制约了其泛化能力的进一步提高.针对此问题,重新定义了原点到超平面的距离,进而提出了基于数据分布信息的单类支撑向量机(DD One-Class SVM).推导构建了DD One-Class SVM算法的优化问题,详细分析和讨论了该优化问题在散度矩阵奇异情况下的求解方法以及该算法的非线情况.相对于传统One-Class SVM算法,该算法体现出了更好的泛化能力.
展开▼
