Graphics Group @ ISU

Forensic firearms examiners subjectively compare two bullets, using visual examination to make a determination about whether the bullets are similar enough to have originated from the same weapon. Recently, there has been increased demand for quantitative, objective similarity assessment methods for many types of forensic evidence, including bullets. In this talk, we discuss visualizations for a machine learning algorithm capable of matching striation marks on fired bullets. We describe the steps of the algorithm and summary visualizations used at each stage of the data science pipeline, and present an application which wraps the visualizations into an interactive exploration tool. This application can be used to explain and interpret results from the machine learning algorithm, but it is also an effective tool for analysis of the model’s strengths and weaknesses when examining new (and sometimes messy) data. Using a sequence of case studies originating from a set of 5 fired bullets, we describe the effectiveness of the application for model exploration. Finally, we evaluate the tool’s design with respect to its ability to bridge the gap between experts in data science and experts in firearms examination.