Tracing the roots of illicit Cannabis: A machine learning-assisted ATR-FTIR proof-of-concept study for geographical origin classification.

Cannabis remains one of the most commonly trafficked and seized illicit drugs around the globe. The growing number of cannabis seizures has created an increasing demand for analytical methods able to determine the geographical origin of seized material rapidly and reliably. Existing approaches are often destructive, require extensive sample preparation, and have limited applicability to Indian cannabis. To address these limitations, the present study explored Fourier-transform infrared spectroscopy in attenuated total reflectance sampling mode (ATR-FTIR) combined with machine learning (ML) approaches for the geographical classification of cannabis samples collected from North India. A total of 200 female cannabis flower-top samples were collected from four distinct geographical regions and 1000 spectral observations were recorded using ATR-FTIR spectroscopy. The spectra obtained presented characteristic absorption bands related to cannabinoids, flavonoids, cellulose, hemicellulose, lignin, and other phytochemical constituents. Then, dataset was reduced using principal component analysis (PCA) and subsequently used to train four ML models. Among the models evaluated, the artificial neural network (ANN) exhibited the best classification performance during the training phase. Its reliability was further validated internally using test data and externally using unknown data. Further, PC3 was the most contributing component to the classification across all models and was associated with holocellulose, lignin, pectin, THCA, and THC. The proposed work has a strong potential for forensic investigations, especially for drug trafficking and identification of cannabis origin in illicit market.