摘要
针对现有基于支持向量数据描述(SVDD)的多类分类算法未能充分利用重叠区域样本分布信息等问题,提出了一种基于核空间相对密度的SVDD多类分类算法DM-SVDD。该算法首先由SVDD确定包围每类数据的最小超球,然后计算位于最小超球重叠区域中每个样本在其同类样本间的相对密度,最后以各类样本相对密度的均值为标准,对重叠区域内的待测样本进行分类。实验结果表明,算法DM-SVDD是可行有效的。
In order to solve the problem that existing multiclass classification algorithm based on support vector data description（SVDD） is not fully use the sample distribution information in overlapping regions.This paper proposed an algorithm for SVDD multiclass classification based on relative density in kernel space （DM-SVDD）.Firstly, DM-SVDD got the minimum bounding hyper-spheres enclosing each class by using SVDD.Then the algorithm calculated relative density with the same class samples for each data in overlapping region of the minimum bounding hyper-spheres.At last,classified the test samples locating in overlapping region according to the mean of relative density of each class. Experimental results show that DM-SVDD is feasible and effective.
出处
《计算机应用研究》
CSCD
北大核心
2010年第5期1694-1696,共3页
Application Research of Computers
基金
江苏省"青蓝工程"资助项目
江苏省六大人才高峰项目(07-E-025)
江苏省高校自然科学重大基金研究项目(08KJA520001)
关键词
支持向量数据描述
多类分类
核空间相对密度
support vector data description（SVDD）
multiclass classification
relative density in kernel space
