Laser induced breakdown spectroscopy (LIBS) aided by machine learning algorithms for milk analysis
Loading...
Date
2024-09-10
Authors
Πλιάτσικα, Νεφέλη
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
In this study, Laser Induced Breakdown Spectroscopy (LIBS) combined with machine learning algorithms was used for the classification of milk samples based on their animal origin, geographical origin, and milking season. Specifically, 1470 raw, unprocessed milk samples of different animal origin (cow, goat, and sheep), originating from three major regions of Greece (Peloponnese, Ionian Islands and Central Greece), were collected during the period 2021-2023 and studied.
First, the plasma emission spectra of all milk samples were recorded, and the atomic emissions corresponding to the main minerals in milk, such as magnesium, calcium, sodium, and potassium, were identified. The emission spectra for the different species, regions and milking season were compared with each other to detect similarities and/or differences. Machine learning algorithms were then applied for the classification of milk LIBS spectra and the construction of predictive models. The algorithms used were Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Logistic Regression (LR), Support Vectors Machines (SVMs) and Gradient Boosting (GB). The algorithmic models LDA, LR, SVMs and GB were applied to the LIBS spectra of milk samples for classification based on animal and geographical origin. PCA was then used as a means of pre-processing the LIBS data; to reduce dimensionality and the models were applied anew to improve performance. In the case of animal origin, the classification and prediction accuracies of the algorithmic models resulted in values of 95-98% and 89-94%, respectively. In the classification analysis based on geographical origin, the classification accuracy reached values of 91% to 98% while the prediction accuracy was estimated between 93% and 96%. Regarding the milking season discrimination analysis, no PCA pre-processing was applied to the spectroscopic data, as exceptionally high classification, and prediction values (∼99%) were achieved without reducing dimensionality.
These results were highly encouraging and clearly demonstrated the effective use of the LIBS technique in combination with machine learning algorithms in the study of milk, especially, for the classification of milk samples based on animal and geographical origin and milking season, which represent areas with limited to no research. This approach indicates the potential of the LIBS technique in the field of food safety and food quality in general.
Description
Keywords
Laser induced breakdown spectroscopy, Machine learning, Milk analysis, Elemental analysis, Classification