|
Abstract:
|
In efforts to strengthen the statistical foundations of forensic evidence interpretation, Bayes Factors and likelihood ratios have been advocated for quantifying the value of evidence. However, both methods rely on formulating convincing statistical models for the evidence, which is particularly challenging for highly complex evidence. Alternatively, score-based likelihood ratios are less challenging from a modeling standpoint. There are many methods of computing comparison scores for score-based likelihood ratios, but many popular methods rely on statistical machine learning algorithms. One of the difficulties with these methods is that the pairwise comparison of all the evidential objects results in a set of dependent scores. The issue lies in the fact that while machine learning methods do not have any distributional assumptions, most assume independence between the observations in the data. Before score-based likelihood ratios of this sort are used in forensic settings, it is important to explore the effects of using this dependent score-data in a machine learning algorithm. We explore these effects using simulated data and real data from casework glass evidence.
|
