摘要
针对现有的多类分类算法效率低下的问题,提出一种K-means聚类算法和超球结合的多类分类算法。对每一类样本,先使用K-means算法获得子类;再在各个子类上构造最小超球,由此对每类都获得一个超球集;这些超球将样本空间分割,根据样本点所在空间的位置综合得到决策函数,用于对输入样本点进行类别判断。从理论上分析该方法能够有效提高分类的速度和准确率。
As current multi-class classification methods are low in efficiency,this paper gave a multi-class classification method based on K-means cluster and hyper-sphere.Firstly it used K-means cluster acquire the baby classes of every father class,then made hyper-sphere of every baby class.After this work,divided the text space up to every different areas.Aim at the position of text and these areas,fabricated decision-making function respectively which could classify a text to the correct class.The analyse given here clearly shows that this method can efficiently improve the speed and veracity of the clssifier.
出处
《计算机应用研究》
CSCD
北大核心
2011年第5期1764-1766,共3页
Application Research of Computers
关键词
K-均值聚类算法
高斯性测度
超球
多类分类
K-means cluster
Gaussian distribution estimation
hyper-sphere
multi-class categorization
