Sorting the mob: using geometric morphometrics and machine learning to differentiate kangaroo postcrania for zooarchaeological applications

Abstract

Taxonomic identification of bone is one of the building blocks of zooarchaeological research into human foraging behaviour. However, it can prove difficult in regions, such as Australia, that have biodiverse taxa that are difficult to differentiate using bone morphology. One such case are the kangaroos and wallabies (macropods), one of the most speciose groups of marsupials whose remains are frequently recovered from Australian archaeological sites. Despite their clear importance to Indigenous economies, little research has been undertaken on how to reliably differentiate the postcranial remains of extant macropods. Here we address this gap by applying three-dimensional geometric morphometrics to describe how the astragalus and calcaneus differs between several large macropod genera. We describe taxonomically diagnostic anatomical attributes between genera and identify the size related (allometric) shape variation that could be mistaken for taxonomic differences. We then compare several machine learning models to demonstrate how these can be applied to geometric morphometric data to statistically classify unknown specimens from palaeozoological contexts with a high degree of accuracy. Our results show that non-linear methods of supervised machine learning outperform classical discriminant function analysis when used on our geometric morphometric data. Statistical classification of palaeozoological specimens has the potential to be a valuable tool, where differentiating skeletal remains of closely related taxa continues to prove challenging.