A robust method to improve the regression accuracy of LIBS data: determination of heavy metal Cu in Tegillarca granosa

Abstract

Tegillarca granosa (T. granosa) is susceptible to contamination by heavy metals, which poses potential health risks for consumers. Laser-induced breakdown spectroscopy (LIBS) combined with the classical partial least squares (PLS) model has shown promise in determining heavy metal concentrations in T. granosa. However, the presence of outliers during calibration can compromise the model's integrity and diminish its predictive capabilities. To address this issue, we propose using a robust method for partial least squares, RSIMPLS, to improve the accuracy of Cu prediction in T. granosa. The RSIMPLS algorithm was employed to analyze and process the high-dimensional LIBS data and utilized diagnostic plots to identify various types of outliers. By selectively eliminating certain outliers, a robust calibration method was achieved. The results showed that LIBS spectroscopy has the potential to predict Cu in T. granosa, with a coefficient of determination (R_p²) of 0.79 and a root mean square error of prediction (RMSEP) of 11.28. RSIMPLS significantly improved the prediction accuracy of Cu concentrations with a 43% decrease in RMSEP compared to the PLS. These findings validated the effectiveness of combining LIBS data with the RSIMPLS algorithm for the prediction of Cu concentrations in T. granosa.