Hierarchical clustering for three-way data

A novel clustering model is presented for three-way data that refer to a set of units on which variables are measured or collected at different occasions. The proposal originates from the CPclus model, where both clusters of units and components for variables and occasions are identified in a k-means based framework. Here we develop a hierarchical variant, called H-CPclus, which is implemented using a divisive approach, where the non-hierarchical model is applied recursively to obtain nested partitions. This allows the results to be displayed in a standard dendrogram fashion.