Bayesian multivariate models for case assessment in dynamic signature cases.

Area of Science:

• Forensic Science
• Biometrics
• Data Science

Background:

• Dynamic signatures, recorded on devices like tablet PCs, generate multivariate data including spatial and temporal information.
• The forensic science community faces challenges in managing and analyzing the large volume of data from dynamic signatures.
• Authenticity verification of dynamic signatures is crucial, similar to static signatures, requiring rigorous forensic examination.

Purpose of the Study:

• To address the limitations of current probabilistic methods for analyzing dynamic signature evidence.
• To highlight the necessity of revising statistical models used in forensic signature analysis.
• To improve the evidential support provided by dynamic signature data in authenticity assessments.

Main Methods:

• Utilizing statistical models, including the Bayes' factor, to discriminate between competing propositions regarding signature authenticity.
• Analyzing multivariate data derived from dynamic signature recordings.
• Critically evaluating existing probabilistic solutions and their underlying assumptions.

Main Results:

• Existing probabilistic solutions for dynamic signature evidence have limitations.
• The independence assumption between questioned and reference signature data is often inappropriate and needs removal.
• Statistical models can be adapted to provide a measure of evidential support for dynamic signatures.

Conclusions:

• Accurate forensic analysis of dynamic signatures requires advanced statistical approaches.
• Removing the independence assumption in statistical models is essential for reliable dynamic signature verification.
• Further development of probabilistic methods is needed to effectively manage and interpret dynamic signature evidence in forensic investigations.