Fentanyl Detection Using Surface Enhanced Raman Spectroscopy with Nearest Neighbour Machine Learning Methods

Abstract

Drug checking services provided by supervised consumption sites can significantly reduce the chance of an overdose for people who use drugs. Research of different drug checking methods has strived to find methods which are economical, portable, non-intrusive, and trustworthy. This work explores the use of machine learning models paired with surface-enhanced Raman spectroscopy which has been identified as a promising approach to drug checking. We propose a novel method which leverages k-Nearest Neighbours to create an explainable supervised feature extraction method. Knowing the target substance allows our model to find regions of interest making it more robust to unseen conditions. Our methods result in models that achieve state-of-the-art performance across all tested datasets including a dataset of street samples collected by a supervised consumption site. This work shows promise in the creation of explainable machine learning models to aid drug checking services at supervised consumption sites.