Clustering of Variables for Mixed Data

Jerome Saracco (1, 2, 3) , Marie Chavent (1, 2)

1 CQFD - Quality control and dynamic reliability
2 IMB - Institut de Mathématiques de Bordeaux
3 ENSC - Ecole Nationale Supérieure de Cognitique

This chapter presents clustering of variables which aim is to lump together strongly related variables. The proposed approach works on a mixed data set, i.e. on a data set which contains numerical variables and categorical variables. Two algorithms of clustering of variables are described: a hierarchical clustering and a k-means type clustering. A brief description of PCAmix method (that is a principal component analysis for mixed data) is provided, since the calculus of the synthetic variables summarizing the obtained clusters of variables is based on this multivariate method. Finally, the R packages {\bf ClustOfVar} and {\bf PCAmixdata} are illustrated on real mixed data. The PCAmix (resp. ClustOfVar) approach is first used for dimension reduction (step1) before standard clustering of the individuals (step 2).

Domaines: Statistiques [stat]

Jerome Saracco, Marie Chavent. Clustering of Variables for Mixed Data. Statistics for Astrophysics: Clustering and Classification, 77, EDP Sciences, pp.91-119, 2016, EAS Publications Series, 978-2-7598-9001-9. ⟨hal-01417442⟩