摘要
K均值算法是一种常用的基于原型的聚类算法。但该算法要求用户随机选择初始质心,使得K均值算法受初始化影响较大。二分K均值算法虽然改善了这个问题,但仍然要求用户指定聚类个数,影响了聚类效果。用层次聚类对二分法进行改进,解决了二分K均值算法受用户指定的聚类个数的影响的问题。并结合Chameleon算法,合并划分过细簇,优化聚类结果。仿真实验证明改进的聚类算法的抱团性和分离性优于二分K均值聚类算法。
K-means algorithm is a kind of commonly used clustering algorithm based on the prototype. But the algorithm requires the user to randomly select initial centre of mass, which makes the K-means algorithm greatly influenced by the initialisation. Although the bisecting K-means algorithm has ameliorated this issue, but it still requires the user to specify clustering number, which impacts clustering effect. We use hierarchical clustering to improve bisecting K-means algorithm, thus solve the problem of impact caused by the bisecting K-means algorithm being affected by the number of clustering the user specified. Moreover, we combine the Chameleon algorithm and unite the clusters being divided too fine and optimise the clustering results. Simulation experiments prove that the unifying nature and separation property of the improved clustering algorithm is better than the bisecting K-means clustering algorithm.
出处
《计算机应用与软件》
CSCD
2015年第2期261-263,277,共4页
Computer Applications and Software
基金
广州科技计划项目(7411655926875)
