Statistical models for the support of forensic fingerprint identifications

Publication Type:
Thesis
Issue Date:
2017
Full metadata record
Files in This Item:
Filename Description Size
01front.pdf168.92 kB
Adobe PDF
02whole.pdf19.66 MB
Adobe PDF
For the majority of the 20th century, the forensic practice of fingerprint identification has had unanimous acceptance as reliable, robust, and admissible evidence. However, a number of forensic commentators have questioned the scientific validity of the current practice of fingerprint identification. Moreover, recent well publicised misidentifications have added concerns with the accuracy and quality assurance processes in practice, while fingerprint practitioners have experienced growing pressure to perform identifications from increasing workload and difficult casework. The application of statistical modelling for fingerprint identification is a scientific methodology that provides a quantification of fingerprint evidence that can alleviate such concerns regarding the scientific foundations of fingerprint identification. Moreover, such statistical models can be used as a supportive tool for fingerprint practitioners who are under operational pressure to accurately assess crime marks against other fingermarks in a timely manner. In this dissertation, two statistical modelling frameworks for different fingerprint identification scenarios are proposed. The first variant is called AFIS-centric models that calculate likelihood ratios and are designed to work with AFIS candidate lists, helping the practitioner to decide between match and close non-match correspondences. Two likelihood ratio measures are proposed, one with the aim of evaluating candidate list members as match or a close non-match, the other providing a weight-of-evidence evaluation. The second model variant called a Person-of-Interest (POI) model is designed for the scenario where a rich collection of fingermarks from the same source finger are available to provide a more thorough evidential assessment. Tailored models of skin distortion are built using samples of the POI's finger, using feature vectors that make use of all of the available spatial information, from which a weight-of-evidence likelihood ratio measure is derived. Experimental results illustrate the effectiveness of the AFIS-centric and POI models as supportive tools for casework. The significance of these research results is threefold. Firstly, the proposed AFIS-centric models illustrate how feature vector based models can focus on match and close non-match populations to provide a statistical measure agnostic of an AFIS scores that can be used for workload reduction purposes through candidate list filtering/reordering and quality assurance within the Analysis-Comparison-Evaluation-Verification (ACE-V) framework. Secondly, the proposed feature vectors add robustness and spatial completeness to the model, resulting in highly accurate models that assess real-world case samples accurately. Lastly, both proposed model variants provide a highly robust and accurate quantitative output in the form of a weight-of-evidence measure that can be used to support expert testimony.
Please use this identifier to cite or link to this item: