Implementing machine learning for the identification and classification of compound and mixtures in portable Raman instruments

Abstract

Portable Raman instruments provide quick, nondestructive analysis of organic and inorganic compounds, making it widely applicable in various disciplines. However, the instrument’s accuracy when analyzing pure, or multiple component mixtures is still an aspect that needs improvement. This study explored machine learning algorithms to classify single compounds, binary, ternary, and quaternary mixtures by the compound name, and the compound’s class, using seized drugs and common diluents as a model. The accuracies were ≥ 93% for most pure, binary mixtures, and quaternary mixtures algorithms. Therefore, incorporating machine learning algorithms in portable instruments, can improve the detection of unknown substances with high accuracies.