Forensic science often operates at the boundary between mathematics, medicine, and the law, particularly in contexts where identifying deceased individuals is both scientifically challenging and ethically imperative. This is increasingly the case in situations linked to crime, migration, humanitarian crises, and social marginalization, where the ability to establish identity becomes not only a technical task but also a matter of justice and human rights. In such scenarios, the role of forensic medicine is to “read” the body: to reconstruct events, detect possible wrongdoing, and, crucially, assign an identity to the deceased.
When remains are poorly preserved—as frequently happens in cases involving migrants—standard identification procedures become significantly more complex. The process typically relies on comparing postmortem data, collected directly from the body (such as tattoos, scars, fingerprints, or DNA), with ante mortem information from missing persons. Before reaching the stage of personal identification, forensic experts first reconstruct a biological profile, estimating characteristics such as sex, age, and biogeographical origin. Only after narrowing down potential matches does the process move toward definitive identification, where the evidential threshold must meet the stringent standard of “beyond reasonable doubt.”
Despite advances in forensic methodologies, identifying individuals based on physical traits alone—especially in the absence of DNA—remains largely qualitative and often lacks a rigorous statistical foundation. This gap motivates a recent PhD project in mathematics at the University of Milan, which seeks to bring probabilistic thinking to the core of forensic anthropological identification. The central idea is to develop a principled statistical framework capable of quantifying how likely it is that a given set of observed traits—such as facial marks, moles, or tattoos—belongs to a specific individual.
The project builds on modern developments in forensic statistics, with a particular emphasis on likelihood-based reasoning and Bayesian inference. By framing identification as a problem of comparing competing hypotheses, one can evaluate the evidential weight of observed traits through likelihood ratios, while also incorporating prior information about the individual or the context. At the same time, alternative approaches such as belief function theory offer tools for handling situations where data are incomplete or imprecise, a common occurrence in forensic casework.
A key mathematical challenge lies in modeling the complex structure of the data. Physical traits are not independent, and their dependencies—especially among facial features—must be carefully accounted for. Moreover, variability across populations, measurement error, and observer subjectivity all contribute to uncertainty, which must be explicitly modeled rather than ignored. Addressing these aspects requires the integration of multivariate statistical methods and probabilistic graphical models, alongside the use of anthropological databases to capture population-level distributions.
The research is conducted by the PhD student Ernesto Maria Greco, supervised by a group of mathematicians (Alessandra Micheletti, Daniela Morale, Stefania Ugolini) in collaboration with the forensic anthropology group led by Cristina Cattaneo at the University of Milan, whose innovative work in this area has recently been recognized with an ERC Advanced Grant. This interdisciplinary setting provides both theoretical challenges and access to real-world data, enabling the validation of proposed models through case studies and simulations.
At a broader level, the project raises fundamental statistical questions. How can the evidential value of anthropological traits be quantified in a principled way? What is the appropriate framework to combine uncertainty, dependence, and population heterogeneity? And perhaps most intriguingly, under what conditions can non-genetic evidence achieve a level of reliability comparable to DNA-based identification?
By addressing these questions, the project aims not only to advance applied statistics, but also to contribute to a domain where mathematical rigor has direct societal impact.