2011
Nick Petraco
Chicago, IL
The purpose of this research is to take steps towards developing standardized methodologies and software that can objectively evaluate tool mark comparison and identification. Fifty-eight primer shear marks on 9mm cartridge cases fired from four Glock model 19 pistols along with fifty striation patterns made in lead with five consecutively manufactured chisels have been collected and analyzed using high-resolution, white light confocal microscopy and modern machine learning techniques. The three-dimensional surface topographies were processed with outlier and form removal before ca cubic spline filter was used to extract all "waviness surfaces" – the essential "line" information familiar to firearm and tool mark examiners. Taking the mean of all profiles that made up each surface summarized the waviness topographies. The mean profiles were then subjected to principal component analysis (PCA) for dimensional reduction and support vector machines (SVM) for profile-gun/chisel associations. Bootstrap resampling was used to estimate identification error rates on a larger data set of assumed similar statistical properties. Conformal prediction theory (CPT) coupled with SVM was used to produce computational based associations with a stated level of statistical confidence corresponding to the standard Nevman-Pearson definition. Suggestions are made for describing and employing the methods exploited here in a courtroom setting. Also, a prototype of a web interface for downloading the data generated and software developed for this project will be described.