关键词: 全局中心, 均值最近点, 谱聚类, 聚类评价指标, 聚类算法

Abstract: Aiming at the problems of high iteration number and long computation time in the clustering process of high dimensional data, an improved clustering algorithm is proposed. The algorithm first uses spectral clustering to reduce the dimension of samples, then selects k data objects with the end to end and the largest distance product as the initial clustering center, in the update process of cluster centers, selects data objects nearest to cluster mean as cluster centers. And other data objects are divided into corresponding clusters according to the minimum distance, iterated iteratively until convergence. The experimental results show that the Rand index, Jaccard coefficient and Adjusted Rand Index of  the new algorithm are better than K-means algorithm and other 3 kinds of improved clustering algorithms. In terms of operational efficiency, the new algorithm has shorter clustering time and fewer iterations.

Key words: global center, mean nearest point, spectral clustering, clustering evaluation index, clustering algorithm