Classification of Cannabinoid Spectra Using Machine Learning

Vibrational spectroscopy, encompassing Raman and Infrared (IR) spectroscopy, is a powerful technique that probes the intrinsic vibrations of a molecule, thus providing a unique chemical signature for that molecule. This information is beneficial to differentiate between two similarly structured molecules since their vibrational fingerprint will be different. In an effort to introduce an automated spectroscopic data analysis tool, we explore different Machine Learning (ML) algorithms to identify the chemical structure from the simulated Raman and IR spectra of 22 similar molecules belonging to the class of cannabinoids. In this study, we investigate the best ML approach by using representative synthetic IR/Raman data obtained from quantum chemical calculations of the selected molecular structures. We account for the experimental variability of the spectra by adding different kinds of noise and backgrounds to the simulated spectra such that they mimic experimental conditions such as fluorescence background as well as Gaussian noise. This methodology is used to setup the database to train the ML algorithms. We report the accuracy of the different ML algorithms and the time taken to process the algorithms in differentiating the cannabinoid varieties.