This research investigates how automated tools used in forensics introduce demographic biases and discusses performance unfairness mitigation strategies.
In the authors’ previous work, they evaluated the impact of demographic differentials in automatic matching of latent fingerprints and incorporated these covariates in the ROC curve. The resulting adjusted ROC curve provided error rates that account for an individual’s demographic information, which is a better measure of the discriminatory capacity compared to the pooled ROC curve. The authors’ ROC regression model was also able to handle continuous covariates such as age as well as discrete covariates such as gender and ethnicity. In the current project, the authors extended the preliminary study carried out on right index latent fingerprints to right thumb instances. They investigated: (i) until which extent demographic differentials vary depending on properties specific to the finger instance (e.g., size of the fingertip) and (ii) the effectiveness of the proposed demographic adjusted ROC to handle unfairness. (Publisher abstract provided)
Downloads
Similar Publications
- Detecting Various Burial Scenarios In A Controlled Setting Using Ground-penetrating Radar And Conductivity
- Extraction of shoe-print patterns from impression evidence using Conditional Random Fields
- Recovery and Detection of Ignitable Liquid Residues from the Substrates by Solid Phase Microextraction - Direct Analysis in Real Time Mass Spectrometry