Ensemble clustering based on weighted co-association matrices: Error bound and convergence properties

Vladimir Berikov, Igor Pestunov

Research output: Contribution to journalArticlepeer-review

27 Citations (Scopus)

Abstract

We consider an approach to ensemble clustering based on weighted co-association matrices, where the weights are determined with some evaluation functions. Using a latent variable model of clustering ensemble, it is proved that, under certain assumptions, the clustering quality is improved with an increase in the ensemble size and the expectation of evaluation function. Analytical dependencies between the ensemble size and quality estimates are derived. Theoretical results are supported with numerical examples using Monte-Carlo modeling and segmentation of a real hyperspectral image under presence of noise channels.

Original languageEnglish
Pages (from-to)427-436
Number of pages10
Journal Pattern Recognition
Volume63
DOIs
Publication statusPublished - 1 Mar 2017

Keywords

  • Cluster validity index
  • Co-association matrix
  • Ensemble size
  • Error bound
  • Hyperspectral image segmentation
  • Latent variable model
  • Weighted clustering ensemble

Fingerprint

Dive into the research topics of 'Ensemble clustering based on weighted co-association matrices: Error bound and convergence properties'. Together they form a unique fingerprint.

Cite this