Rapid and cost-effective presumptive screening of firearm discharge-consistent metals via bromopyrogallol red colorimetry and machine learning.

The identification of gunshot residues (GSR) is pivotal in forensic chemistry; however, standard confirmatory techniques, such as Scanning Electron Microscopy with Energy Dispersive X-ray Spectroscopy (SEM/EDS), are often costly, time-consuming, and centralized in major urban facilities. This study proposes a rapid and cost-effective presumptive screening framework for firearm discharge-consistent metals (Pb2 + and Sb3+), integrating Bromopyrogallol Red (BPR) colorimetry, digital image processing, multivariate chemometrics, and machine learning algorithms. One hundred samples were evaluated, including real hand swabs from shooters firing pistols, revolvers, and rifles, negative control samples ("non-shot"), and inorganic metallic standards. Digital images obtained in a standardized lighting environment were processed to extract quantitative chromatic descriptors across RGB and HSV/HSI color spaces. Exploratory analyses via Principal Component Analysis (PCA) and Non-Metric Multidimensional Scaling (NMDS) demonstrated robust class separation (PC1 + PC2 = 83.1%; NMDS stress = 0.027), with luminosity and intensity variables (V, I, R, and L) acting as primary discriminants. Supervised machine learning classifiers were developed to differentiate "shot" from "non-shot" samples, where tree-based algorithms-specifically Random Forest and XGBoost-achieved exceptional discriminatory performance, both yielding an Area Under the ROC Curve (AUC) of 0.998. Notably, the Random Forest model exhibited a minimal false-negative rate and superior probabilistic calibration, critical attributes for forensic decision-making. Executed in under 10 min per sample, this methodology provides a reliable, decentralized triaging solution to filter negative cases locally and prioritize high-probability samples for confirmatory SEM/EDS analysis.