Bayesian Statistics in Forensic Sciences: Probabilistic Models and Application in the Evaluation of Dactyloscopic Evidence

This thesis analyzes the intersection between the scientific method and judicial logic, addressing the management of uncertainty inherent in the reconstruction of the fact in criminal proceedings. The first chapter outlines the theoretical foundations of scientific evidence, with particular attention to the Bayesian approach and the use of the Likelihood Ratio (LR) to measure the discriminant strength of findings when comparing opposing hypotheses, clearly distinguishing the role of the expert from that of the judge. In the second chapter, the theoretical model is applied to forensic fingerprinting: starting from the biological pillars of impression permanence and selectivity, the ACE-V protocol is examined and the transition from categorical certainty judgments to probabilistic quantification of evidence using feature-based and score-based models is described. Finally, the third chapter delves into overcoming deterministic criteria —such as minutiae counting— by developing discrete probabilistic models (binomial, multinomial, Poisson) and conjugate distributions to integrate parameter uncertainty. The discussion concludes by highlighting the main logical fallacies in judicial practice and reiterating how the adoption of an explicit probabilistic formalism can help protect due process.