|
| 摘要: |
| 数据聚类是一个功能强大的技术，它能够把数据特征相似的对象划分为一类，但是并不是所有的聚类算法的实现都能产生相同的聚类结果；并且K均值算法的结果很大程度上依赖它的初始中心的选择；提出了一种新颖的关于K均值初始中心选择的策略；该算法是基于反向最近邻（RNN）搜索，检索一个给定的数据集，其最近的邻居是一个给定的查询点中的所有点；使用这种方法计算初始聚类中心结果发现是非常接近聚类算法所需的迭代聚类中心；对提出的算法应用到K均值聚类中给予了证明；用几种流行的数据集的实验结果表明了该算法的优点。 |
| 关键词: 聚类 最近邻查询 反向最近邻搜索 K均值 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| A Kind of Improvement for K-means Algorithm |
|
LI Guang-ming,LI Liang,ZHANG Jian-gang
|
| Abstract: |
| Data clustering is a powerful technology and can calssify the objects with similar data characteristics into a class,however,the implementation of all clustering algorithms does not produce the same clustering results,moreover,the results of K-means algorithm largely depend on the selection of initial clustering center.This paper proposes a novel strategy about K-means initial clustering center selection,whose algorithm is based on reverse nearest neighbor search and retrieves a given data set whose nearest neighbor is all point in a given inquiry point.The result by using this algorithm to t=calculate initial clustering center reveals that this center is very close to iterative clustering center needed by clustering algorithm.This paper also verifies the application of the proposed algorithm to K-means cluster and uses the experiment through several popular data sets to demonstrate the advantages of this algorithm. |
| Key words: cluster nearest neighbor inquiry reverse nearest neighbor search K-means value |
