2020/02/21 by Nathaniel Garton, Danica Ommen, Danica M. Ommen +6
Biochemistry, Genetics and Molecular Biology · Computer Science · Mathematics · Social Sciences · #Anomaly Detection Techniques and Applications #Applications (stat.AP) #FOS: Computer and information sciences #Forensic Fingerprint Detection Methods #Forensic and Genetic Research #stat.AP
paper · pdf · doi:10.48550/arxiv.2002.09470
openalex publication_date 2020/02/21 · arxiv created 2020/05/22 · arxiv updated 2020/05/25 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
In 2016, the European Network of Forensic Science Institutes (ENFSI) published guidelines for the evaluation, interpretation and reporting of scientific evidence. In the guidelines, ENFSI endorsed the use of the likelihood ratio (LR) as a means to represent the probative value of most types of evidence. While computing the value of a LR is practical in several forensic disciplines, calculating an LR for pattern evidence such as fingerprints, firearm and other toolmarks is particularly challenging because standard statistical approaches are not applicable. Recent research suggests that machine learning algorithms can summarize a potentially large set of features into a single score which can then be used to quantify the similarity between pattern samples. It is then possible to compute a score-based likelihood ratio (SLR) and obtain an approximation to the value of the evidence, but research has shown that the SLR can be quite different from the LR not only in size but also in direction. We provide theoretical and empirical arguments that under reasonable assumptions, the SLR can be a practical tool for forensic evaluations.