Evaluating cross-sectional geometry-based methodologies on an archaeological and historical sample

The construction of a complete biological profile for unknown individuals from skeletal remains plays a key role in the holistic study of archaeological sites. However, the presence of commingled elements, as in the case of mass graves, hinders this process. This work aims to evaluate the diaphyseal cross-sectional geometric properties of the long bones for pair-matching and for sex estimation based on the femur, tibia, and humerus bones on archaeological and historical samples. The sample consisted of 3D virtual bone models of 95 individuals (46 males, 39 females, 10 of unknown sex) from both Greek and English assemblages, utilizing the dedicated “csg-toolkit” GNU Octave package for variable extraction. The pair-matching sorting algorithm performed adequately in all three assemblages, excluding over 97% of the true mismatched pairs. The accuracy rate for certain pair detection was close to 50% for the lower limbs, and 41% for the humerus, with the algorithm deeming a significant amount of the missed pairs as plausible, resulting in 70% to 80% of total pairs to be detected in some capacity for all bone assemblages. The sex estimation classifiers performed sufficiently in all cases, yielding moderately high accuracy rates in the majority of examined samples, with values consistently higher than 70%. However, the application of contemporary methods on archaeological or historical material needs to be performed with caution, as secular changes and population-specific trends can affect the applicability of the methods. As such, the utilization of different techniques before reaching a conclusion is vital and should be encouraged.