The authors previously demonstrated in a research project that electrochemical screening of gunshot residue (GSR) samples delivers a simple, inexpensive, and sensitive analytical solution that is capable of detecting both inorganic and organic gunshot residues (IGSR and OGSR) in less than 10 min. per sample; and in the current study, the previous work is expanded by increasing the number of GSR markers and applying machine learning classifiers to the interpretation of a larger population data set.
Using bare screen printed carbon electrodes, the detection and resolution of seven markers (IGSR; lead, antimony, and copper, and OGSR; nitroglycerin, 2,4 dinitrotoluene, diphenylamine, and ethyl centralite) were achieved with limits of detection (LODs) below 1 µg/mL. A large population data set was obtained from 395 authentic shooter samples and 350 background samples. Various statistical methods and machine learning algorithms, including critical thresholds (CT), naïve Bayes (NB), logistic regression (LR), and neural networks (NN), were used to calculate the performance and error rates. Neural networks proved to be the best predictor when assessing the dichotomous question of detection of GSR on the hands of shooter versus nonshooter groups. Accuracies for the studied population were 81.8 percent (CT), 88.1percent (NB), 94.7 percent (LR), and 95.4 percent (NN), respectively. The ability to detect both IGSR and OGSR simultaneously provides a selective testing platform for gunshot residues that can provide a powerful field testing technique and assist with decisions in case management. (publisher abstract modified)
Downloads
Similar Publications
- Single-cell Investigative Genetics: Single-cell Data Produces Genotype Distributions Concentrated at the True Genotype Across All Mixture Complexities
- Microbiome Data Accurately Predicts the Postmortem Interval Using Machine Learning Regression Models
- Fraud Victimization Across the Lifespan: Evidence on Repeat Victimization Using Perpetrator Data