Classification of roasted coffee bean products using laser-induced breakdown spectroscopy: a novel variable selection approach for multiclass modeling

Abstract

The classification of coffee beans by species, origin, and quality is essential in the coffee industry to ensure authenticity and consistency. While existing methods like spectroscopic and chromatographic techniques offer valuable insights, some require complex sample preparation, while others, such as near-infrared (NIR) and visible/near-infrared (VIS/NIR), rely on molecular information that is labile during coffee roasting. Laser-induced breakdown spectroscopy (LIBS), a fast and minimally invasive elemental analysis technique, shows promise for food authentication. In this study, we evaluated the feasibility of combining LIBS with the k-nearest neighbors (k-NN) algorithm to classify 12 roasted coffee bean products available in South Korean markets. LIBS spectra revealed emission peaks for elements such as Li, Na, K, Rb, Mg, Ca, C, H, and O, along with molecular emission bands of CN and C₂. Using the newly developed statistical concept of the ‘inter-to-intraclass variation ratio,’ the emission intensities of Li, Na, and Rb were identified as key discriminatory variables for the classification model. The k-NN model achieved a classification accuracy of 96.0% with k = 1, which improved to 98.5% with standard deviation-based scaling and k = 3. It should be emphasized that the model based on the Li, Na, and Rb composition is not expected to be labile during the coffee bean roasting process. These findings underscore the potential of LIBS, combined with a simple machine-learning algorithm, as a practical and efficient tool for authenticating coffee products, leveraging its high sensitivity to alkali metal elements for rapid and accurate classification.