Testing the Wits Facial Classification Scheme (WFCS) for forensic facial comparison in black African males

Abstract

The crime rate increase in South Africa calls for new approaches to fighting crime and ensuring that perpetrators and victims are identified with little to no error using scientific evidence-based approaches. Forensic facial comparison (FFC) is an emerging field with huge potential to aid in identification of individuals in the age of widespread surveillance. The current Facial Identification Scientific Working Group (FISWG) feature list used in facial comparison analyses is a descriptive list that provides a list of morphological features for the identification of unknown individuals. This list was previously used to test the validity of morphological analysis in FFC. From these studies, a revised facial feature classification scheme called the Wits Facial Classification Scheme (WFCS) was developed, based on the FISWG feature list. The aim of this study was to test the performance of the WFCS within a black male South African sample. The WFCS entailed scoring of 412 morphological facial features individually for each face. The sample size consisted of 100 face pools (50 WT and 50 CCTV). The number of total facial comparisons was 119 (63 for WT and 56 for CCTV). Average kappa agreement scores and composite facial comparison scores were used to evaluate the performance of the WFCS under two distinct image capture conditions: standard CCTV images and wildtype optimal photographs (WT). Statistical analyses were conducted using ROC curves, confusion matrices, Mann-Whitney U, and Kruskal-Wallis ANOVA by ranks tests. The results showed a high repeatability (95%) and reproducibility (73%) and better performance of the WFCS under the WT condition than in the CCTV. Balanced accuracy was higher in the WT condition for both the average kappa agreement (92%) and composite facial comparison scores (76%). In the CCTV condition, balanced accuracy was 80% for the average kappa agreement and 61% for the composite facial comparison scores. The predictive quality values revealed that overall, the WFCS demonstrated superior performance when assessed using average kappa agreement scores compared to composite facial comparison scores. The FISWG feature list performed better than the WFCS in terms of balanced accuracies. The WFCS should be refined and tested on different populations and genders as it is still a more statistically testable method of identification. Implementing a more statistically testable method, may improve the legal defensibility of forensic facial comparison.