PCS-granularity weighted ensemble clustering via Co-association matrix

Ensemble clustering has attracted much attention for its robustness and effectiveness compared to single clustering. As one of the representative methods, most co-association matrix-based ensemble clustering typically only take into account a single type of information contained in base partitions. This study proposes a new weighted ensemble clustering algorithm of fusing multi-level data information to sufficiently mine the information from the base partition family. Three different levels of data information, including partition granularity level, cluster granularity level and sample granularity level, are concomitantly considered in the co-association matrix. More specifically, we utilize knowledge granularity to measure the quality of base partitions, and rough membership to quantify the credibility of base clusters; Additionally, the relative similarity of a pair of samples is estimated with respect to different base partitions, taking into account the close relationship between samples and the structure of base clusters. Subsequently, the partition-cluster-sample-granularity weighted co-association (PCSCA) matrix is proposed to address the limitations of the co-association matrix, quantifying the quality of information at multiple levels. Finally, this study introduces the partition-cluster-sample-granularity weighted ensemble clustering (PCSEC), which incorporates the PCSCA matrix. The experimental results demonstrate the effectiveness of the proposed method.