Abstract
Ensuring milk quality and detecting adulteration are among the critical issues for maintaining regulatory compliance and for consumers’ health and safety. In the present work, Laser-Induced Breakdown Spectroscopy (LIBS) is used for the rapid and in situ discrimination/classification of commercial milk samples originating from the major dairy product manufacturing companies in Greece. The milk samples are discriminated/classified based on animal origin, fat content, and bio/non-bio labeling. To analyze the LIBS spectroscopic data, three machine learning algorithms were evaluated: Linear Discriminant Analysis (LDA), Support Vector Machines (SVM), and Logistic Regression (LR). While these models achieved satisfactory accuracy in classifying milk samples based on animal origin and fat content, the discrimination between bio- and non-bio-labeled milk samples proved more challenging. In addition, the detection and quantification of adulteration/mixing of goat milk with cow milk are also examined. The spectroscopic data, analyzed by Partial Least Squares Regression (PLSR), yielded very good results, reaching high validation metrics (R2 = 0.96, RMSE = 6.86), and successfully quantifying the degree of adulteration/mixing of goat milk with cow milk. The present results highlight the efficacy of integrating LIBS and machine learning for rapid milk characterization. Utilizing a single spectroscopic technique to classify samples and quantify adulteration presents practical applications in quality control, product authentication, and compositional assessment.