Advancing automobile identification and brand discrimination from tyre rubber through Machine learning algorithms for forensic investigations

Abstract

Criminal instances involving collision accidents, hit-and-run incidents, abduction, hostage-taking, and the unauthorised transit of forbidden items generally include evidence involving rubber traces from automobile tyres. These traces can be located on the road surface, in clothing, on the victim(s) themselves, or on items as skid marks following sudden stopping and spinning around. These traces serve as crucial evidence by reducing the range of suspects by revealing linkages between the getaway vehicle, the site of the crime, and the perpetrator through the tyre's brand, producer, or origin. This study offered a way for classifying 220 tyre rubber samples from different brands using various machine learning algorithms in PyCaret in conjunction with rapid and non-destructive ATR-FTIR spectroscopy equipped with diamond crystal. On spectral information from ATR-FTIR, pre-processing tools such as baseline correction, smoothing, derivatization, and normalisation were also implemented prior to machine learning. This approach has the potential to be advantageous for efficiently and non-destructively identifying rubber traces as forensic evidence and for facilitating brand recognition of automobile tyres.