基于投票机制的融合聚类算法

以一趟聚类算法作为划分数据的基本算法,讨论聚类融合问题.通过重复使用一趟聚类算法划分数据,并随机选择阈值和数据输入顺序,得到不同的聚类结果,将这些聚类结果映射为模式间的关联矩阵,在关联矩阵上使用投票机制获得最终的数据划分.在真实数据集和人造数据集上检验了提出的聚类融合算法,并与相关聚类算法进行了对比,实验结果表明,文中提出的算法是有效可行的.

Custer Fusion Algorithm Based on Majority Voting Mechanism

Taking the one-pass clustering algorithm as the basic algorithm for grouping data, the issue of clustering ensemble is investigated. Over multiple clusters obtained by random threshold and sequence of data input of the one-pass clustering algorithm, produces a mapping of the clusters into an association matrix between patterns. The final data partition is obtained by voting mechanism over this association matrix. Experimental results of the proposed cluster fusion algorithm on several real and synthetic data sets are compared with clustering results produced by well known clustering algorithms. The experimental results show that the proposed algorithm is effective and practicable.