Abstract
Data cubes are p-dimensional data structures (p ≥ 3). This study proposes alternative clustering strategies for four-dimensional data cubes (p = 4) with independent slices. In biological sciences, the key point of Statistical-Data Analysis is the study of variability focusing on its structure and components, while interactions between slices and hyper-slices are often of scientific interest. The dimensionality reduction strategy decomposes the cube's total variability into variability between and within slices and hyper-slices. The first clustering strategy projects the extracted principal components as points in Euclidean space, the second projects them onto a differentiable manifold, and the third is based on Procrustes distances among matrices. The real dataset “Basil” is used, consisting of three irrigation levels (hyper-slices), five basil varieties (slices), seven quantitative traits, and 60 basil plants. These strategies highlight the need for a holistic approach to the cube, so that its underlying sources of variability can be studied in depth and from different perspectives.