Clustering per le matrici a tre vie

Abstract

Consider N statistical units (U), p variables (X), and K occasions or sources of data (O). Cluster analysis for three way and three mode data matrices can concern itself with many aspects. We can group the K matrices in m homogeneous clusters following the scheme { O; (U; X) } . In the scheme { U; (X; O) } we seek m clusters of matrices (variables per occasion) of homogeneous units. In the scheme { X; (U; O) } the matrices (units per occasion) are grouped in function of homogeneous variables. For two mode and three way matrices the schemes are {O; (X; X)} and {O; (U; U) }. In this paper, after recalling the INDCLUS and ADCLUS algorithms, two more recent proposals by Bellacicco-Ronzoni and Chiodi are illustrated. At the end a new algorithm, PRINCLUS, based on the individuation of m < K principal matrices is presented.