A Bayesian decision approach to classify crime scene observations in sharp-force fatalities: a study on suicide vs. homicide scenarios

A Bayesian decision approach is proposed to address complex discrimination problems in forensic medicine. From a judicial perspective, the accurate assessment of cases of violent death as a suicide or a homicide is of undeniable importance for criminal prosecution, guiding decisions on the depth of forensic investigation and the requirement for additional analytical procedures. Inaccurate decisions can have severe consequences; therefore, a normative perspective is introduced to provide a framework through which current practice can be articulated. This study represents a proof of principle based on a dataset of 51 cases (24 suicides and 27 homicides). A Bayesian logistic regression model is implemented to assign a probability value to the event that a questioned scenario belongs to a given population (e.g., suicide or homicide). However, the computation of such probability and the subsequent decision-making process remain a demanding task. Probabilistic graphical models, specifically Bayesian Networks (BNs), offer a suitable and user-friendly framework to assist forensic pathologists in providing statistically sound expert opinions. The implementation of the proposed BN facilitates the application of this approach in routine forensic practice.