Date of Graduation
2018
Document Type
Thesis
Degree Type
MS
College
Eberly College of Arts and Sciences
Department
Forensic and Investigative Science
Committee Chair
Keith Morris
Committee Co-Chair
James Carroll
Committee Member
Casey Jelsema
Abstract
The results of a data set of five models of 9 mm Luger caliber handguns suggest an algorithm such as the square of the Mahalanobis Distance as a step towards the creation of an electronic class characteristic database for cartridge cases to supplement the currently existing General Rifling Characteristic (GRC) database for bullets. The algorithm was validated using both hold-one out cross validation as well as on an entirely independent set of ground truth known cartridge cases which were blindly classified to mimic case work. A method for determining an objective threshold for inclusion onto a list for an investigator is proposed and the effects are illustrated using the blind set of hypothetical case work. The algorithm relied upon quantitative measurements of class characteristics. Three firearms per model and ten test fires per firearm were used to inform the algorithm of the mean and variance of the measurements taken. The test fires used to inform the algorithm were a combination of physical test fires and previous IBISRTM entries where ground truth model was known. Measurements were taken of images which were retained in a digital cloud filing system where folders were used to organize test fires by the known donor model and firearm. Hold-one out cross validation was performed by withholding the measurements for a given test fire to serve as a questioned cartridge case, and computing the Mahalanobis Distance to each model. A threshold cut-off distance for inclusion on to a list which would be provided to an investigator was calculated based upon the results of the hold-one out cross validation and based upon the known-match Mahalanobis Distances following a central Chi-Square distribution. This threshold cut-off distance was used to guide decisions during the blind and independent classification of individual, physical cartridge cases. Each blind cartridge case was classified one at a time, independent of GRC bullet information, to mimic a crime scene where only one cartridge case is recovered. The blind set also included cartridge cases originating from models outside of the five considered by the algorithm or "database," representative of real challenges experienced in case work.
Recommended Citation
Wang, Young, "Inclusion of 9 mm Firearm Type Using Quantitative Class Characteristics" (2018). Graduate Theses, Dissertations, and Problem Reports. 6910.
https://researchrepository.wvu.edu/etd/6910