关键词: 集成聚类, 聚类质量, 群体一致性, 信息熵, 一致函数

Abstract: Aiming at the problem that existing ensemble clustering algorithms usually used K-means algorithm as the base clustering generator, although it could ensure the diversity of clustering members, it ignored that poor base clusterings might cause terrible disturbance to the final clustering result, we proposed a two stage ensemble algorithm based on clustering quality. Considering that K-means algorithm ran efficiently, but the clustering quality was relatively rough, firstly, we proposed to  use K-means algorithm to generate base clustering members in the generation stage, and then  selected clustering members with both high quality and strong diversity through  group aggrement measure to form candidate ensemble. Secondly, the information entropy knowledge was futher applied to construct the weighted-clustering co-association matrix in the ensemble stage. Finally, the final clustering result was obtained by using consensus function. Three indexes were used for comparative experiments on ten real datasets, and the experimantal results show that the algorithm can effectively improve the accuracy of clustering results while maintaining good robustness.

Key words: ensemble clustering, clustering quality, group aggrement, information entropy, consensus function