Speaker:  Hana Lee, PhD Candidate, Department of Statistics, Iowa State University

Title: Classification-Based Matching Rules With Application in Image Data for Forensic Problems

Abstract:  We propose nonparametric matching rules that leverage classification to predict if two items of interest originated from the same source. A classifier is employed to predict the likelihood of each item belonging to a set of known sources. Subsequently, the outcomes of the two items are used to quantify the scores indicating their common origin. This approach utilizes both classification outcomes and their associated uncertainty. Additionally, we show a mathematical relationship between classification and our matching rule in a closed-set framework. Then, we implement the proposed matching rule on various image and numerical data and demonstrate its out-performance at small false positive rates compared to several existing methods.