Evaluation of Forensic Data Using Logistic Regression-Based Classification Methods and an R Shiny Implementation

An overview of log likelihood ratio cost in forensic science - Where is it used and what values can we expect?

There is increasing support for reporting evidential strength as a likelihood ratio (LR) and increasing interest in (semi-)automated LR systems. The log-likelihood ratio cost (Cllr) is a popular metric for such systems, penalizing misleading LRs further from 1 more. Cllr = 0 indicates perfection while Cllr = 1 indicates an uninformative system. However, beyond this, what constitutes a "good" Cllr is unclear. Aiming to provide handles on when a Cllr is "good", we studied 136 publications on (semi-)automated LR systems. Results show Cllr use heavily depends on the field, e.g., being absent in DNA analysis. Despite more publications on automated LR systems over time, the proportion reporting Cllr remains stable. Noticeably, Cllr values lack clear patterns and depend on the area, analysis and dataset. As LR systems become more prevalent, comparing them becomes crucial. This is hampered by different studies using different datasets. We advocate using public benchmark datasets to advance the field.

A methodological comparison of discriminant function analysis and binary logistic regression for estimating sex in forensic research and case-work

The purpose of this study is to assess the accuracy of two multivariate statistical approaches for estimating sex from human external ear anthropometry, namely, discriminant function analysis (DFA) and binary logistic regression (BLR). A cross-sectional sample of 497 participants (233 males and 264 females) aged 18-35 years (24.42 ± 5.17) was obtained from Himachal Pradesh state of North India. Both the ears of the participants (994) were examined for anthropometric measurements. A total of 12 anthropometric measurements were taken independently on the left and right ear of each individual with the help of a pair of sliding calipers using a standard method. The sex of the population groups was discriminated against using binary logistic regression and discriminant function analysis. The predictive percentage of sex estimation computed from both the models were substantially the same, that is, 76.3% from DFA and 76.2% from BLR, with nearly comparable (∼0.02) sensitivity, specificity, positive predictive value, and negative predictive values, whereas the values of correct predicted percentage were 0.1% higher in DFA than BLR. Moreover, the other comparison metrics, such as classification error, B-index, and Matthews correlation coefficient indicated that both models performed equally well. The study highlighted that if the assumptions of the statistical methods are met, both methods are equally capable of discriminating the population depending on sex. The study recommends that the discriminant function analysis and binary logistic regression may be used synonymously in forensic research and case-work pertaining to the estimation of sex and various other forensic situations.

Assessing significant factors that can influence physical fit examinations - Part I. Physical fits of torn and cut duct tapes

This study expands upon a previously developed method that quantifies the similarity of the compared tape edges by systematically studying the effect of several separation methods and tape grades on the quality of a fit. Analysts examined more than 3300 pairs of hand-torn or scissor-cut duct tapes from three different tape grades while they were kept blind from the ground truth to minimize bias. The samples were examined following a three-step methodology: 1) qualitative assessment of the overall edge alignment and description of edge pattern, 2) macroscopic evaluation of the edges' features, 3) bin by bin subunit assessment of tape edges and estimation of the edge similarity score. A report template was designed to maintain records of the decision-making process. In the second and third steps, eight comparison features were defined and documented using auto-populated cell options. Generally, misidentification rates were low, with no false positives reported. Coinciding with previous research, low scores (under 20%) provided the most support for a non-fit conclusion, while high scores (80% or higher) supported a fit conclusion. A statistical analysis of the separation method and quality of tape revealed a potential interaction between these factors and showed that they significantly impact the edge scores for true fitting pairs, but not the true non-fits' scores. The developed comparison and documentation criteria can assist practitioners with a more straightforward, consistent, and transparent interpretation and reporting approach.