Most globular clusters in the Milky Way show signs of multiple stellar subpopulations. These may be due to distinct star formation events, as suggested by differing chemical abundance patterns. Photometric coordinates designed to be sensitive to light-element abundances -the so-called chromosome-map plane- were recently introduced to distinguish subpopulations without resorting to spectroscopy. However, drawing boundaries between subpopulations on the chromosome-map plane is somewhat subjective, as it ultimately depends on the judgment of a human expert. I will discuss unsupervised machine learning strategies to make this process automatic so that it can become reproducible and be generalised to a large set of GC observations. In particular I will show that the AGglomerative NESting (AGNES) algorithm with Ward's linkage produces results similar to expert expectations on NGC 2808.