Bayesian models in questioned handwriting and signatures


Ressource 1Download: thèse-GL-OK.pdf (7560.49 [Ko])
State: Public
Version: After imprimatur
License: Not specified
Serval ID
PhD thesis: a PhD thesis.
Bayesian models in questioned handwriting and signatures
Gaborini Lorenzo
Taroni Franco
Institution details
Université de Lausanne, Faculté de droit, des sciences criminelles et d'administration publique
Publication state
Issued date
Forensic document examination is one of the oldest areas of forensic science.
Despite the advent of personal computers and portable digital tools, the discipline has enjoyed relatively few methodological advances compared to other forensic areas.
Moreover, the use of handwritten evidence in court has historically faced many issues, particularly in the US-centric system.
Among the specificities of this field, one can identify the lack of physical laws governing the fundamental principles of handwriting, the difficulty in assessing the general validity of these principles, and the reliance on human experts' judgement to provide an opinion to the stakeholders.
The reporting of the evidential value also lags behind other forensic areas, such as DNA interpretation, where the modern Bayesian approach of the ENFSI Guideline for Evaluative Reporting in Forensic Science is fully implemented.
Starting from the fundamental principles governing handwriting, necessarily qualitative in their nature, this thesis first considers several well-defined forensic scenarios in which such principles can be translated to a statistical description of a series of measurements.
The necessary evidence is collected either on request via a panel of writers, or from real casework.
We considered scenarios where authorship is discussed, either of signatures or naturally handwritten content.
Next, the Bayesian approach is introduced, from the theoretical notions to the computational requirements.
As no universal technique to compute the Bayes Factor is available (the only coherent measure for evaluative purpose), every scenario requires a distinct path and a tailored approach.
The stability and the validity of each developed model is also approached, for instance by performing sensitivity analyses on its parameters or on the data.
Bayesian reasoning can be easily generalized, and allows one to approach the issue of combining multiple kinds of evidence.
As an example, we consider a hypothetical scenario where an anonymous letter is found jointly with salivary evidence, under the hypothesis that both came from the same person of interest.
The person of interest declares that his twin brother was the source of both traces.
In the last Chapter we first show how the developed models can be adapted for each type of evidence, then how they can be combined together to produce an evaluative report that is coherent and justified.
bayesian statistics, forensic handwriting examination, signatures, Bayes factor, evidence
Create date
28/11/2023 20:27
Last modification date
15/12/2023 8:57
Usage data