Decoding the Past: An Uncertainty-Aware Deep Learning Framework for Sex Attribution in Prehistoric Hand Stencils
Determining the biological sex of the individuals who created Upper Paleolithic hand stencils remains challenging due to the absence of ground truth, morphological differences between contemporary and prehistoric populations, and uncertainties arising from image degradation. Traditional morphometric methods are further constrained by substantial overlap between sexes, limited cross-population generalizability, and subjective feature selection. This study introduces a deep learning framework that explicitly models, propagates, and aggregates uncertainty throughout the analytical pipeline. As a proof of concept, each stencil is represented by twelve plausible silhouette realizations to account for boundary uncertainty, which are processed by two ensembles of ten deep neural networks trained on contemporary hand samples to learn potentially transferable morphological patterns. A triangulated validation scheme integrates ensemble predictions, two-dimensional latent-space analysis, and spatial attribution maps to assess the consistency of the results from complementary perspectives. Applied to prehistoric hand stencils, the framework produces sex predictions alongside confidence measures based on internal agreement, helping distinguish morphologically consistent cases from ambiguous ones. The combined evidence from ensemble predictions, latent-space organization, and spatial attribution demonstrates the potential of explicitly quantifying uncertainty in archaeological inference, providing a systematic and interpretable approach to the analysis of prehistoric hand art.
Authors
- Ramón A. Mollineda (ORCID: https://orcid.org/0000-0001-5765-2905)
- Dean Snow (ORCID: https://orcid.org/0000-0003-0824-618X)
- Boris Mederos (ORCID: https://orcid.org/0000-0002-0131-7566)
- Karel Becerra (ORCID: https://orcid.org/0009-0007-7658-7142)
Institutions
- Universitat Jaume I (ES)
- Universidad Autónoma de Ciudad Juárez (MX)
- Pennsylvania State University (US)
Publication Details
- Journal
- Machine Learning and Knowledge Extraction
- Published
- 2026-11-16
- DOI
- https://doi.org/10.3390/make8110382
- Primary Topic
- Forensic Anthropology and Bioarchaeology Studies
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Agencia Estatal de Investigación