Nonparametric Bayesian Clustering Ensembles

Forming consensus clusters from multiple input clusterings can improve accuracy and robustness. Current clustering ensemble methods require specifying the number of consensus clusters. A poor choice can lead to under or over fitting. This paper proposes a nonparametric Bayesian clustering ensemble (NBCE) method, which can discover the number of clusters in the consensus clustering. Three inference methods are considered: collapsed Gibbs sampling, variational Bayesian inference, and collapsed variational Bayesian inference. Comparison of NBCE with several other algorithms demonstrates its versatility and superior stability.

• Type: Journal
• Year: 2010
• Where: PKDD
• Authors: Pu Wang, Carlotta Domeniconi, Kathryn Blackmond Laskey
• Keywords: Clustering Ensemble, Consensus Clusters, Data Mining, PKDD 2010, Variational Bayesian Inference