基于支持向量数据描述算法的SVM多分类新方法

New multi-class support vector machine based on support vector data description

提出一种基于支持向量数据描述算法(SVDD)的多分类方法(S-MSVM)。受SVDD的启发,该方法对每类样本建立一个超球来界定,但训练好的超球在所有情况下都是相交的。选择相交区域的样本单独建立超球,重复该步骤,直到相交区域消失或相交区域内没有样本点。给出了该方法的时间复杂度分析,并通过实验验证了该方法具有相对较好的训练精度。

This paper proposed a method of multi-class problem based on SVDD. Impired from SVDD, this method constituted a hyperspherical classifier for the sample of each class. But the hyperspheres which had been trained well were sharing a common region in every case. To solve this disadvantage, the sample in intersectant region chosed to constitute the hyperspheres, Repeated this process until intersectant region disappear or there was no sample in intersectant region. This paper presented the computational complexity of this method, and proved its higher training precision.