摘要
提出一种改进的近邻新聚类算法,该算法具有较高的效率,只需计算一遍样本间的广义距离,即可完成初步的聚类。该算法最大特点是:在很大程度上使聚类结果避免陷入局部解;不用预知类别就可对大批数据进行分类,并能指出可能的异常数据。仿真结果证明该算法大大优于LBG法[1]和模糊聚类法[5]
This paper puts forward a improved nearest neighbor clustering algorithm. This algorithm is very effident' It can obtain a primary solution that is near to the global optimal solution only by calculating the distant between the data for one time. This algorithm can not only avoid local lutions in most case, but also classify large quantities of data without knowing what the classes should be beforehand, and point out which datum is probably abnormal.The computer emulating result proved that this algorithm excel the LBG and fuzzy clustering algorithm.
出处
《山东建材学院学报》
1999年第2期122-124,共3页
Journal of Shandong Institute of Building Materials
关键词
聚类算法
聚类中心
势力圈
歪
总歪
clustering
cluster centroid
innuence circle
deviation, overall deviation