Sparse Divisive Feature Clustering

Ndèye Niang-Keita (1) , Mory Ouattara (1) , Gilbert Saporta (1)

1 CEDRIC - MSDMA - CEDRIC. Méthodes statistiques de data-mining et apprentissage

We propose an approach based on a divisive algorithm for clustering variables in order to identify in a large data table underlying dimensions that are not necessarily orthogonal. The number of clusters does not have to be defined in advance. The clusters, which are as unidimensional as possible, are then represented in a parsimonious way by a small number of variables or components.

Keywords: Feature Clustering; Simple structure; Sparse principal component

Domains: Mathematics [math]; Statistics [math.ST]; Computer Science [cs]; Machine Learning [cs.LG]

Ndèye Niang-Keita, Mory Ouattara, Gilbert Saporta. Sparse Divisive Feature Clustering. XXVIII Meeting of the Portuguese Association for Classification and Data Analysis (JOCLAD 2021), CLAD, Dec 2021, Covilhã, Portugal. pp.75-76. ⟨hal-03475860⟩