聚类分片双支持向量域分类器

针对支持向量域分类器对大规模样本集的训练时间长且占用内存大的问题，构造聚类分片双支持向量域分类器。以均值聚类剖分原始空间，并选取密度指标大的样本作为初始聚类中心；对子空间构造双支持向量域分类器，根据样本与正负类最小包围超球的距离构造分段决策函数；定义样本的变尺度距离，以链接规则组合子空间的分类结果。数值实验表明，所提出算法的分类精度高且受参数变化的影响不大，分类时间短且随子空间数的增加而降低。

Clustering piecewise double support vector domain classifier

Support vector domain classifiers have disadvantages like long training time and large memory. The clustering piecewise double support vector domain classifier(CPDSVDC) is proposed. CPDSVDC uses C means algorithm to partition the original space, and selects the initial cluster centers by samples with large density indexes. The dual support vector domain classifier is constructed in each divided subspace, and the corresponding piecewise decision function is also constructed based on the position relationship between the test sample and the two minimum enclosing spheres. The variable distance of the test sample is defined, and linking rule is used to combine classification results in all subspaces. Numerical experiments demonstrate that the CPDSVDC has high classification accuracy that varies slightly with parameters and low training time that decreases with the number of subspaces.