Sparse and group-sparse clustering for mixed data An illustration of the vimpclust package
CHAVENT, Marie
Université de Bordeaux [UB]
Méthodes avancées d’apprentissage statistique et de contrôle [ASTRAL]
Voir plus >
Université de Bordeaux [UB]
Méthodes avancées d’apprentissage statistique et de contrôle [ASTRAL]
CHAVENT, Marie
Université de Bordeaux [UB]
Méthodes avancées d’apprentissage statistique et de contrôle [ASTRAL]
< Réduire
Université de Bordeaux [UB]
Méthodes avancées d’apprentissage statistique et de contrôle [ASTRAL]
Langue
en
Communication dans un congrès
Ce document a été publié dans
JDS 2022 - 53èmes Journées de Statistique de la Société Française de Statistique (SFdS), 2022-06-13, Lyon.
Résumé en anglais
High-dimensional data may often contain both numerical and categorical features, and in some cases features may be available as natural groups (repeated measurements, categories of features, ...). Clustering this kind of ...Lire la suite >
High-dimensional data may often contain both numerical and categorical features, and in some cases features may be available as natural groups (repeated measurements, categories of features, ...). Clustering this kind of data raises several issues: how to simultaneously deal with numerical and categorical features? how to build meaningful clusters of the input entities? how to select the most informative features or groups of features for the clustering? In the k-means framework, one may rely on a penalised version of the between-cluster variance, and find both the best partitioning of the data, and the most informative features or groups of features. The present manuscript illustrates sparse k-means and group sparse k-means for mixed data, using the vimpclust package. The example provided on a small real-life dataset shows how feature selection may be directly combined with clustering, and provide a meaningful selection while preserving the quality of the clustering.< Réduire
Mots clés
clustering
k-means parcimonieux
pénalités L1 et L1-groupe
données mixtes
packages R
Mots clés en anglais
clustering
sparse k-means
L1 and group-L1 penalties
mixed data
R packages
Origine
Importé de halUnités de recherche