ABSTRACT
Cluster analysis aims to form homogeneous groups of multivariate samples, called clusters, such that samples with similar sets of values are placed in the same cluster, and the clusters are as different as possible in terms of their constituent members. Two crucial decisions underlie any clustering algorithm: (1) how to define a measure of distance between the samples, which quantifies appropriately how “close” samples are to one another; and (2) how to define a corresponding measure of distance between the clusters. The special nature of compositional data and the distance functions discussed in previous chapters will prescribe the appropriate measures to use. Clustering can also be performed on the compositional parts themselves.
