Machine learning assisted SERS detection of selenium species using a sodium alginate/silver hydrogel substrate.

Selenium (Se) was an essential trace element that possessed a very narrow concentration range between nutritional benefit and toxic effects. The development of rapid and accurate detection methods for selenium speciation in environmental samples was therefore very important. In this study, a portable, stable, and highly sensitive sensing platform was developed for rapid qualitative and quantitative analysis of selenium speciation in complex water and soil samples. Sodium alginate/silver nanoparticles (SA/Ag NPs) hydrogel substrates for surface-enhanced Raman scattering (SERS) were successfully synthesized by a simple Ca2+ ion crosslinking method. The SA/Ag NPs hydrogel substrate showed excellent uniformity, reproducibility, and stability, with an enhancement factor (EF) reached 109. The detection limits of the SA/Ag NPs substrate for Se(IV) and Se(VI) were 3.35 μg/L and 3.37 μg/L, respectively. Density functional theory (DFT) calculation explored the SERS chemical enhancement mechanism. The performances of classification models such as PLS-DA, PCA-DA, KNN, and SVM, and regression models including PLSR and SVR were systematically evaluated for identifying and quantifying selenium speciation in real water and soil samples. The SVM model achieved an accuracy of 90.0%, the coefficient of determination (R2) of SVR was greater than 0.8958, and the root mean square error (RMSE) was less than 54.6 μg/L. This method established a reliable framework integration scheme of a hydrogel SERS substrate and a machine learning algorithm, which has good practical application potential.