APPROACHES TO LIKELIHOOD RATIO ESTIMATION FOR FORENSIC EVIDENCE INTERPRETATION

Forensic science is a critical element of the criminal justice system. Forensic scientistsexamine and analyze evidence from crime scenes and elsewhere to develop objective findings that can assist in the investigation and prosecution of perpetrators of crime or absolve an innocent person from suspicion. Biometric traits such as DNA, faces, and fingerprints play an important role in forensic science. Fingerprint examination is one of the oldest specialty areas within the forensic sciences. In 2002, Joseph L. Peterson and Matthew J. Hickman analyzed fingerprint examinations. In this study, they compared fingerprints of unknown origin found at crime scenes with known fingerprints in a database. The fingerprint analysis was once viewed as an error-free process. However, the performance of fingerprint examination also depends on the experience of the examiner and individual judgment, which may cause the results not to be accurate; an innocent person could be wrongly accused, so the accuracy of forensic examiners is crucial. The wrong identification may cause very serious consequences for individuals and society. Hence, it is critical to assess the accuracy of the forensic examination. This dissertation is aimed at the development of advanced statistical methods quantifying the strength of forensic evidence based on the score-based likelihood ratio. In the first chapter of this dissertation, we introduce essential background about forensic evidence interpretation, and introduce important statistical quantities relevant to this task. In the second chapter of this dissertation, we present a collection of methods for estimating likelihood ratios (or density ratios), and investigate the performance of these methods in simulations and real data analysis involving fingerprint analysis and facial recognition. In the third chapter of this dissertation, we discuss extensions of those approaches to adjust for covariates observed along with biometric or forensic measurements.

Author

• Qi, He

Identifier

hdl:1920/14776

OAI identifier oai:identifier

oai:MARS:1920/14776