New Forensic Study Uses Machine Learning to Predict TNT Residue Recovery After Explosions
Researchers use machine learning and LC-MS/MS to predict TNT residue recovery after explosions, examining how surfaces and sampling conditions affect results.
Researchers in Türkiye have developed a machine learning approach to predict how much TNT residue forensic investigators may recover from different surfaces following an explosion. The study, published on November 5, 2026, in the Journal of Forensic Sciences, examines how sampling conditions influence the detection of explosive residues.
Recovering chemical traces after an explosion is an important part of forensic investigation. However, the quantity of residue collected can depend on the material being examined, its distance from the explosion, and the sampling method used.
Even sensitive laboratory instruments cannot detect substances that were not successfully collected at the scene.
To investigate these challenges, researchers conducted controlled TNT detonations at a secured testing location with bomb disposal specialists and crime scene investigators.
They positioned metal, wood, and polyester surfaces at distances ranging from one to three metres from the blast centre.
Investigators then collected residues using swabs pre-wetted with methanol, isopropyl alcohol, or acetonitrile.
The collected samples were examined using liquid chromatography–tandem mass spectrometry (LC–MS/MS), an analytical technique capable of separating and detecting chemical compounds.
The researchers compared how different surface materials, solvents, and distances affected the relative amount of TNT recovered.
Their results showed that metal surfaces generally produced stronger and more consistent TNT signals than wood or polyester.
Acetonitrile typically demonstrated higher extraction efficiency, although its performance varied depending on the surface material.
As expected, recoverable TNT decreased significantly as the distance from the explosion increased.
The team also developed a supervised machine learning model using Support Vector Regression (SVR) to examine relationships between the experimental variables.
The model successfully reproduced the observed recovery trends, including nonlinear relationships between surface properties, sampling conditions, and TNT detection.
These findings suggest that predictive models could eventually help investigators make more informed decisions about where and how to collect explosive residues.
The research is particularly relevant because post-blast evidence collection often takes place under difficult conditions, where investigators must prioritize sampling locations and preserve potentially limited chemical evidence.
However, the model was developed using controlled experimental data involving TNT and a defined range of surfaces, solvents, and distances.
Its reported performance does not establish that it can reliably predict residue recovery across different explosive formulations, environmental conditions, or operational crime scenes.
Additional validation would be necessary before adopting the approach as a routine forensic decision-support tool.
The study highlights how combining analytical chemistry with statistical modeling may improve understanding of evidence recovery, while reinforcing the importance of scientifically validated sampling procedures.