On the Number of Clusters in Block Clustering Algorithms

Malika Charrad (1) , Yves Lechevallier (2) , Mohamed Ben Ahmed (1) , Gilbert Saporta (3)

1 ENSI - École Nationale des Sciences de l'Informatique [Manouba]
2 AxIS - Usage-centered design, analysis and improvement of information systems
3 CEDRIC - MSDMA - CEDRIC. Méthodes statistiques de data-mining et apprentissage

One of the major problems in clustering is the need of specifying the optimal number of clusters in some clustering algorithms. Some block clustering algorithms suffer from the same limitation that the number of clusters needs to be specified by a human user. This problem has been subject of wide research. Numerous indices were proposed in order to find reasonable number of clusters. In this paper, we aim to extend the use of these indices to block clustering algorithms. Therefore, an examination of some indices for determining the number of clusters in CROKI2 algorithm is conducted on synthetic data sets. The purpose of the paper is to test the performance and ability of some indices to detect the proper number of clusters on rows and columns partitions obtained by a block clustering algorithm.

Mots clés: Block Clustering; Machine learning; Number of Clusters; Biclustering; Validity Indices

Mots clés: Classification double; Indices de validité; Nombre de Clusters; Bloc de Clustering; Apprentissage automatique

Domaines: Informatique [cs]; Statistiques [math.ST]

Malika Charrad, Yves Lechevallier, Mohamed Ben Ahmed, Gilbert Saporta. On the Number of Clusters in Block Clustering Algorithms. 23rd International FLAIRS Conference, May 2010, Florida, United States. pp.392-397. ⟨hal-01125839⟩