Assessing the Feasibility of Mid- and Near-Infrared Spectroscopy for Classifying Plant-Based Beverages and Predicting Nutrient Content

The increasing demand for plant-based beverages as alternatives to dairy milk requires rapid and reliable methods to assess their composition and authenticity. This study investigated the feasibility of mid-infrared (MIR) and near-infrared (NIR) spectroscopy for classifying plant-based beverages and predicting their nutritional profile. A total of 57 commercial beverages from five categories (oat, almond, soybean, rice, and coconut) were analyzed. Canonical discriminant analysis was used to discriminate among beverage categories, and modified partial least-squares regression models were developed using reference chemical analyses and spectral data to predict protein, fat, sugars, ash, acidity traits, minerals, and amino acid composition. Both MIR and NIR successfully discriminated among the five beverage categories. Quantitative prediction models were generally more accurate with MIR than with NIR. The most robust MIR models were obtained for protein, fat, glucose, ash, and selected minerals (P and K), reaching accuracy levels suitable for quality-control applications. Protein was predicted with excellent accuracy by both technologies. The most robust prediction models were achieved for amino acid composition, with all amino acids except phenylalanine showing satisfactory predictive performance, particularly with MIR spectroscopy. In conclusion, these results demonstrate the feasibility of infrared spectroscopy for the authentication and compositional assessment of plant-based beverages. In particular, MIR spectroscopy showed considerable potential for integration into routine quality-control workflows, similar to those currently implemented for dairy milk analysis.