This research explores the clustering ensemble problem, which tries to aggregate various base clustering to generate performance that is superior to that of the individual one. As a weighted linear combination of the connective matrices from various base clustering, the existing clustering ensemble methods typically create a co-association matrix, which indicates the pairwise similarity between samples. The resulting Co-association matrix is then adopted as the input of a pre-existing clustering algorithm, such as Meta clustering. The co-association matrix, meanwhile, could be dominated by weak base clustering, leading to weak performance. In order to address the issue, we suggest a new matrix of similar label approximation based approach in this study. We specifically create a Cohesive Matrix, which comprises a small but highly reliable set of links between samples, by examining whether two samples are grouped to the same cluster with various base clustering. The Cohesive Matrix and the Co-association matrix are then stacked to create a three-dimensional Matrix, whose label correspondence quality is further investigated to convey the Cohesive Matrix's information to the Co-association matrix and create a more accurate co-association matrix. We frame and effectively solve the proposed approach as a smooth confined optimal solution. Comparing the proposed model to 11 state-of-the-art approaches, experimental results over 7 benchmark data sets demonstrate that it delivers a breakthrough in clustering performance. To the best of our knowledge, this study to investigate the potential of a clustering ensemble using a matrix of similar label, which is fundamentally different from other methods. Last but not least, our technique just has one easily adjustable parameter.

#