摘要
在进行聚类分析的时候,许多算法需要事先给出聚类类别数目。如果在不了解原始数据内部结构的情况下,很难给出合适的聚类类别数目。因此,通过对目前的聚类算法的研究,提出了一种聚类类别数目自动学习算法。算法在分析原始数据的内在分布的基础上,通过统计分布测试,能够准确的判断数据分裂的合理性,最终得到能充分拟合原始数据的聚类类别数目。实验证明算法是可行并且是有效的。
Class number for clustering a dataset is the precondition of the classical algorithm. It is very difficult to confirm the appropriate class number until the original internal structure of dataset is caught on. By researching currently clustering algorithm, a new automatically clustering algorithm is proposed, It can estimate the rationality of data division precisely by evaluating the internal distributing of original data and taking statistics distribution test. Finally, the conclusion that class number is consistent with original data is obtained. By experimeuts,, the new algorithm is proved to be feasible and effective.
出处
《计算机工程与设计》
CSCD
北大核心
2007年第2期252-253,256,共3页
Computer Engineering and Design
关键词
聚类
类别数目
自动学习
假设检验
统计分布
clustering
class number
automatically learning
hypothesis testing
statistics distribution