Estimation of Score-Based Likelihood Ratios in the Presence of Covariates, with Application to Forensics and Biometric Recognition

• Activity Number: 336 - Statistical Modeling and Machine Learning for National Security Applications
• Type: Contributed
• Date/Time: Tuesday, August 9, 2022 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Defense and National Security
• Author(s): He Qi* and Martin Slawski and Larry Tang
• Companies: George Mason University and George Mason University and University of Central Florida
• Keywords: Forensic; Score-based likelihood ratios; Univariate density ratios; Covariates; Semi-parametric location-scale model; Facial recognition

Abstract

The specific source problem in forensic pattern interpretation entails the comparison of evidence from a crime scene to the corresponding piece of evidence from a suspect. The observed agreement with that specific source is compared to the agreement with alternative sources from a background population in terms of a likelihood (or density) ratio, large values of which are considered confirmatory of the hypothesis that the evidence from the crime scene matches the suspect. In this talk, we present a semi-parametric location-scale model to estimate univariate density ratios (commonly referred to as score-based likelihood ratios in forensics) in the presence of covariates of interest. Examples of such covariates include subjects' demographics, years of experience of forensic examiners, and measurement characteristics (e.g., image quality) that are known to affect the discriminative in biometric recognition problems. The proposed approach is evaluated on a facial recognition study.