Research On Likelihood Ratio Evidence Evaluation Method Based On Fingerprint Matching Scores

The increasing demand for scientific nature of forensic evidence in litigation has highlighted the importance of the scientific validity of fingerprint evidence.The main issue is that due to the difficulty in empirically demonstrating the uniqueness of fingerprints,there are foundational validity flaws.Additionally,the current fingerprint identification mainly relies on the subjective experience of the examiner without scientific quantification of fingerprint features.The identification conclusions made do not have uniform objective standards,leading to certain deficiencies in the applied validity of fingerprint evidence,which in turn has led to recurring exposure of judicial errors related to fingerprint evidence,severely impacting its efficacy.Furthermore,in China,fingerprint identification conclusions have traditionally used expressions such as "Identification","Exclusion" or "Inconclusive" limiting the effective utilization of fingerprint evidence.Conducting an objective assessment of factors influencing fingerprint identification,scientifically quantifying the strength of fingerprint evidence,and establishing a fingerprint evidence evaluation system that conforms to the logic of evidence proof is key to addressing the scientific validity issues of fingerprint identification and maximizing the utilization of fingerprint evidence.Many experts and scholars both domestically and internationally are increasingly supporting the use of statistical models to construct scientific evaluation methods for fingerprint evidence.Among them,the likelihood ratio model is considered by the forensic science community as the most correct and logical paradigm for evaluating forensic evidence,and it has gradually become a research hotspot.This thesis focuses on the specific research of constructing,validating,and applying a likelihood ratio evaluation model of fingerprint evidence based on matching scores,aiming to promote the transition of fingerprint evidence evaluation methods from empirical to scientific.The core research content and innovations are as follows:1.Constructed same-source and different-source fingerprint databases and utilized an automatic fingerprint identification system as a quantitative tool to evaluate factors influencing fingerprint evidence assessment and obtain matching scores for constructing a likelihood ratio evaluation model.The same-source fingerprint database obtained various deformed fingerprints through the method of "live collection of deformed fingerprints + recording screenshots",addressing the challenges of obtaining a large number of same-source fingerprint samples and comprehensively simulating deformation methods.The different-source fingerprint database originated from a large-scale fingerprint database used in policing practice,featuring abundant sample quantities,reasonable structures,and meeting the requirements for stable data distribution patterns.2.Established likelihood ratio evaluation models for different minutiae areas and quantities,determined the optimal distribution pattern of fingerprint matching scores using parameter methods,and resolved the issue of not being able to obtain all matching scores after querying different-source fingerprint databases with millions of samples through moment estimation and maximum likelihood estimation methods,concluding that the optimal distribution pattern for all matching scores of different-source fingerprints is the Beta distribution.The validation scheme for the likelihood ratio evaluation model has been designed.Validation results indicate that the likelihood ratio model under different numbers of minutiae performs relatively better than models based on different areas.The likelihood ratio model for the fingerprint center region exhibits better performance metrics compared to models for the other two areas,which have relatively higher false positive and false negative rates exceeding 10%,posing a certain risk of results misleading.As the number of minutiae increases,the discriminative ability of the likelihood ratio model gradually strengthens.With 9-12 minutiae,the false positive rate is approximately 2%,decreasing to around 1% starting from 13 minutiae,indicating a lower risk of result misinterpretation,and aligning closely with qualitative analysis results in current fingerprint identification practices.3.Proposed a method of combining different minutiae areas and quantities into a highdimensional dataset of 36 dimensions,using principal component analysis to reduce data dimensionality and construct a multi-dimensional likelihood ratio evaluation model.The use of principal component analysis for data dimension reduction addresses the issue in parameter methods where different minutiae areas and quantities are modeled separately without considering various minutiae areas and quantities comprehensively.The constructed twodimensional likelihood ratio evaluation model demonstrates excellent discriminatory capabilities,essentially achieving complete differentiation between same-source and differentsource fingerprints.4.Developed likelihood ratio value magnitude intervals and language scales to explain the relative strength of support under same-source and different-source hypothesis conditions,proposed recommendations for reporting fingerprint evidence evaluation results using likelihood ratio values,and emphasized the prerequisite of correctly interpreting likelihood ratio calculation results and effectively avoiding reasoning fallacies,which involves measuring the likelihood ratio values and their corresponding relative support strengths under specific case circumstances after assessing the competing claims of same-source and different-source hypotheses via likelihood ratio evaluation methods.This thesis lays a solid foundation for academic research and practical application in constructing a scientific fingerprint evidence evaluation system.It also provides valuable references for the examination and identification of theories and practical research of other impression evidence with morphological features,to enhance the scientific validity of forensic methods and the utility value of forensic evidence,thereby advancing the adoption and factual determination of morphology evidence.