一种基于引力的分层聚类算法

摘要

Thc traditional hierarchical clustering algorithm for clustering process, only uses the distance between samples as the sole criterion for similarity, this description is too simple. Associated with the formation of galaxies in the universe is essentially a clustering process by gravitational attraction between galaxies role. Introduce the idea of hierarchical gravitational clustering, propose a hierarchical clustering algorithm based on gravitational HCBG ( Hierarchical Clustering Base Gravity), from two aspects of the distance between the samples and the cluster size classes more accurately depicts the similarity. The hierarchical clustering process is regarded as the sample points based on "gravity" to attract spontaneous process. Use UCI machine learning datahase: Iris, Wine nnd Glass as data sets, experimental results show that the proposed algorithm HCBG clustering results than classical hierarchical clustering based on distance HC ( Hierarchical Clustering) increase 5％ ～ 10％ or so.%传统的分层聚类算法在聚类过程中,仅使用样本间的距离作为相似度的唯一标准,其描述过于单一.考虑到宇宙中星系的形成过程本质也是一种聚类过程.星系之间吸引力是靠万有引力作用.将万有引力思想引入分层聚类中,提出一种基于引力的层次聚类算法HCBG(Hierarchical Clustering Base Gravity),从样本间的距离和类簇的大小两个方面更加精确地刻画相似度.把分层聚类的过程看成样本点之间依据"万有引力"自发吸引的过程.采用UCI机器学习数据库的Iris,Wine和Glass数据集,实验结果表明,提出的HCBG算法的聚类结果比经典的基于距离的层次聚类HC(Hierarchical Clustering)提高5%～10%左右.