一种改进的自适应聚类集成选择方法

针对自适应聚类集成选择方法（Adaptive cluster ensemble selection，ACES）存在聚类集体稳定性判定方法不客观和聚类成员选择方法不够合理的问题，提出了一种改进的自适应聚类集成选择方法（Improved ACES，IACES）.IACES依据聚类集体的整体平均归一化互信息值判定聚类集体稳定性，若稳定则选择具有较高质量和适中差异性的聚类成员，否则选择质量较高的聚类成员.在多组基准数据集上的实验结果验证了IACES方法的有效性:1）IACES能够准确判定聚类集体的稳定性，而ACES会将某些不稳定的聚类集体误判为稳定；2）与其他聚类成员选择方法相比，根据IACES选择聚类成员进行集成在绝大部分情况下都获得了更佳的聚类结果，在所有数据集上都获得了更优的平均聚类结果.

An Improved Adaptive Cluster Ensemble Selection Approach

Adaptive cluster ensemble selection (ACES) is not only non-objective in judging the stability of cluster ensemble but also unreasonable in selecting cluster members. To overcome such drawbacks, an improved adaptive cluster ensemble selection (IACES) approach is proposed. First, IACES judges the stablity of cluster ensemble according to its total average normalized mutual information. Second, if cluster ensemble is stable, then cluster members with high quality and moderate diversity are selected, else, cluster members with high quality are selected. We evaluate the proposed method on several benchmark datasets and the results show that IACES can judge the stability of cluster ensemble correctedly while ACES misjudges some unstable cluster ensemble as stable. Besides, ensembling the cluster members selected by IACES produces better final solutions than other cluster member selection methods in almost all cases, and is superior average results in all cases.