Skip to content

Ensemble of SLR Systems for Forensic Evidence

Conference/Workshop:
American Academy of Forensic Sciences (AAFS)
Published: 2022
Primary Author: Federico Veneri
Secondary Authors: Danica Ommen

Score-based likelihood ratios (SLR) have been used as an alternative to feature-based likelihood ratios when the construction of a probabilistic model becomes challenging or infeasible [1]. Although SLR has been shown to provide an alternative way to present a numeric assessment of evidential strength, there are still concerns regarding their use in a forensic setting [2]. The SLR approach requires two key components. First, developing a (dis)similarity score, and second, estimating the distribution of the scores under both prosecutor and defense propositions. This process relies on the construction of pairwise comparisons in both stages. Previous work addresses the dependency on the sets used for producing SLR, how introducing perturbation to the sets can lead the forensic examiner to different conclusions, particularly the sensitivity during the second stage [3]. A second less explored dependence structure is produced because forensic glass evidence can be thought to be generated by a hierarchical model [4]. When a pairwise comparison is made, the same source enters the comparison multiple times, undermining the independence assumption often required by the methods used for developing a (dis)similarity score and estimating their distribution. We introduce an ensemble method for creating an SLR system that aims to reduce sensitivity to perturbation in the samples and remove the dependence structure. We create independent subsets by sampling only one comparison from each source to create base SLR systems, and then ensemble them in a final system that can be used in a later stage.

Related Resources

Error Rate Methods for Forensic Handwriting Identification

Error Rate Methods for Forensic Handwriting Identification

Presentation is from the 106th International Association for Identification (IAI) Annual Educational Conference
Measuring Proficiency among Latent Print Examiners: A Statistical Approach from Standardized Testing

Measuring Proficiency among Latent Print Examiners: A Statistical Approach from Standardized Testing

This presentation is from the 74th Annual Scientific Conference of the American Academy of Forensic Sciences
Ensemble of SLR systems for forensic evidence

Ensemble of SLR systems for forensic evidence

Machine learning-based Score Likelihood Ratios have been proposed as an alternative to traditional Likelihood Ratio and Bayes Factor to quantify the value of forensic evidence. Scores allow formulating comparisons using…
Likelihood Ratios for Categorical Evidence with Applications to Digital Forensics

Likelihood Ratios for Categorical Evidence with Applications to Digital Forensics

In forensic investigations, the goal of evidence evaluation is often to address source-/identity-based questions in which the evidence consists of two sets of observations: one from an unknown source tied…