Application of Machine Learning to Toolmarks: Statistically Based Methods for Impression Pattern Comparisons

• NCJ Number: 239048
• Author(s): Nicholas D. K. Petraco, Ph.D.; Helen Chan, B.A.; Peter R. De Forest, D.Crim.; Peter Diaczuk, M.S.; Carol Gambino, M.S.; James Hamby, Ph.D.; Frani L. Kammerman, M.S.; Brooke W. Kammrath, M.A., M.S.; Thomas A. Kubic, M.S.,J.D., Ph.D.; Loretta Kuo M.S.; Patrick McLaughlin; Gerard Petillo B.A.; Nicholas Petraco, M.S.; Elizabeth W. Phelps, M.S.; Peter A. Pizzola, Ph.D.; Dale K. Purcell, M.S.; Peter Shenkin, Ph.D.
• Date Published: March 2012
• Length: 99 pages
• Grant Number(s): 2009-DN-BX-K041
• Agencies: NIJ-Sponsored
• Publication Type: Report (Study/Research)
• Award: Application of Machine Learning to Toolmarks: Statistically Based Methods for Impression Pattern Comparisons

Annotation

This project's goal was to provide a scientific basis for the reliability and validity of impression evidence, specifically impressions made by tools and firearms, by laying down, testing, and publishing methodological statistical foundations for toolmark impression pattern recognition and comparison.

Abstract

The study focused on striation patterns left by tools and on cartridge casings from firearms. Since all impressions made by tools and firearms can be viewed as mathematical patterns composed of features, this study used the mathematics of multivariate statistical analysis in order to recognize variations in these patterns. In the context of computational pattern recognition, this is called "machine learning." The mathematical details of machine learning can yield what Moran calls "...the quantitative difference between an identification and non-identification" (Moran 2002). Mathematical details also enable the estimation of extrapolated identification error rates and, in some case, the calculation of rigorous, universal random-match probabilities. The current project was divided into three main tasks. First, toolmark pattern collection and archiving was conducted. Second, database and Web interface were constructed for the distribution of toolmark data, accompanied by related software development. Third, multivariate machine-learning methods relevant to the analysis of collected toolmarks were identified and used. This research succeeded in composing a set of objective and testable methods for associating toolmark impression evidence with the tools and firearms that produced them. Three-dimensional confocal microscopy, surface metrology, and multivariate statistical method are the core of the approach presented. 59 figures, 1 table, and 89 references