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

Singh M.; Sharma A.; Sharma V. · Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy · 2026

**Background & Methods** This proof-of-concept study developed a non-destructive analytical method to classify cannabis geographical origin using attenuated total reflectance Fourier-transform infrared spectroscopy (ATR-FTIR) combined with machine learning. A total of 200 female cannabis flower-top samples were collected from four distinct geographical regions in North India, yielding 1000 spectral observations that were analyzed using principal component analysis (PCA) for dimensionality reduction followed by training of four machine learning classification models. **Key Findings** • Artificial neural network (ANN) demonstrated superior classification performance compared to other machine learning models evaluated, with reliable internal validation using test data and external validation using unknown samples for geographical origin discrimination. • ATR-FTIR spectral analysis identified characteristic absorption bands corresponding to cannabinoids (THCA and THC), flavonoids, cellulose, hemicellulose, lignin, and other phytochemical constituents that differentiate cannabis samples by geographical origin. • Principal component 3 (PC3) emerged as the most significant contributor to geographical classification across all models and was primarily associated with holocellulose, lignin, pectin, THCA, and THC content variation between regions. **Dosage & Administration** Not reported. **Safety & Adverse Effects** Not reported. **Evidence Quality** This is a preliminary proof-of-concept study with significant limitations. The study is confined to North Indian cannabis samples, restricting geographical generalizability to other cannabis-producing regions globally. The sample size of 200 flowers, while providing 1000 spectral observations, remains relatively modest for robust machine learning model development. The methodology relies on spectroscopic and chemometric techniques without independent validation against established gold-standard methods. As a 2026 publication in a spectroscopy journal, peer review status and reproducibility by independent laboratories remain unknown. The study lacks comparison with existing destructive and non-destructive analytical approaches. Despite these limitations, the work represents an innovative application of ATR-FTIR spectroscopy combined with machine learning for forensic drug analysis with potential practical applications in law enforcement and drug trafficking investigations.

More recent cannabis research