Sorting of Proximal and Distal Long Bone Fragments Using Diaphyseal Cross‐Sectional Geometric Properties

ABSTRACT The re‐association of skeletal remains has been one of the core challenges for anthropological studies of commingled skeletal elements. However, there has been a lack of methodologies focusing on the re‐association of fragmented remains. This work proposes a novel method for the screening of fragmented long bone diaphysis, based on the geometric properties of their 20% and 80% diaphyseal cross‐sections and utilizing support vector machine models. The sample consisted of 3D long bone models from 396 femora, 422 tibiae, and 415 humeri from the modern Athens Collection. The proposed algorithm was further evaluated on an archaeological and historical sample of Greek and British ancestry. The yielded true negative rate values on the reference sample were promising, ranging from 30% (femur) to over 60% (humerus sample). The total accuracy values, calculated for the top five most probable pairs, followed a similar trend (≈30% for humeri, ≈16% for femora). In all three bone sets, however, the rates were notably higher than the respective random expectation (calculated as 5 over the number of possible pairs). The evaluation on the archaeological and historical dataset showed that the algorithm performed generally better on the tibiae, while further examination of the results indicated that its performance differed based on the sample size. Based on these results, this approach has potential as a screening method for commingled assemblages, contributing to narrowing down the potential matches in larger assemblages. Furthermore, it fills the existing gap in literature, as there is a lack of similar methods in the field. However, further optimization and evaluation on different samples is important, as well as other follow‐up analyses.