摘要
基于归并聚类中心的思想,将全部样本作为初始聚类中心,以离差隶属度作为计算聚类中心的因素,用最大类间距离作为归并聚类中心的标准,进而确定出聚类的数目和最终聚类中心,得出聚类结果。通过实验数据的验证表明,本方法得出的聚类结果能够有效的反映出待聚类样本的真实情况,并且与待聚类样本的初始顺序无关,同时具有一定的抗噪能力。
In many cluster analysis methods, k-means cluster of classical statistics has been widely used because of its fast quality. However, it needs the given cluster numbers, the cluster results are deferent which depend on the sample initial sequence and are easy to be influenced by outliers. It may cause supervised, not real, sequencing and not exact results. According to this problem, this paper, based on center-merger thinking, put all samples to be initial centers, use deviation and threshold to calculate the finial cluster results. After using this method and comparing with the k - means cluster method on the experimental data, it shows its validity of cluster analysis.
出处
《航空计算技术》
2007年第4期64-66,共3页
Aeronautical Computing Technique
关键词
聚类分析
模糊统计
模糊聚类
归并聚类中心
cluster analysis
fuzzy statistic
fuzzy cluster analysis
merger cluster center