Quantitative determination of phosphorus in seafood using laser-induced breakdown spectroscopy combined with machine learning

Abstract Quantitative determination of phosphates or total phosphorus in seafood is of great importance for the fraud detection as well as food security issues. In this work, laser-induced breakdown spectroscopy (LIBS) was applied as a rapid method for phosphorus determination in three types of seafood including codfish, scallop and shrimp. Both univariate and multivariate regression models were established with special attentions on the correction of matrix effect to improve the analytical performances of LIBS. The obtained results showed that compared with the traditional univariate model and the linear PLS model, the non-linear SVM model could provide the best figures-of-merit with R2 of 0.9904, RMSEC, RMSEP and ARE of 1.68 g/kg, 1.42 g/kg and 3.70%, respectively. The average RSD of prediction of SVM is 5.18%, which is much lower than the value of PLS (9.40%) and is comparable to the value of univariate model (4.11%). This indicates that SVM may be more suitable to address the non-linear behaviors in LIBS spectra caused by the matrix effect, and therefore leads to a more robust calibration model. The present results demonstrated the capacity of LIBS combined with machine learning in phosphorus determination of seafood products, which could be potentially used for on-site phosphates detection within the food supply chains.