Quantitative analysis of toxic elements in polypropylene (PP) via laser-induced breakdown spectroscopy (LIBS) coupled with random forest regression based on variable importance (VI-RFR)

With the extensive use of plastic products, determining the amount of toxic elements in plastic products has become a pressing issue for human health and environmental protection. A combination of laser-induced breakdown spectroscopy (LIBS) and analytical methods has been used for the quantitative analysis of plastics, such as calibration methods and calibration-free methods. However, these methods present low precision in the quantitative analysis of plastic, and they do not focus on the determination of toxic elements (Cr and Hg), which are also harmful for human health and cause environmental problems. Chemometrics, a new multidisciplinary branch of chemistry, can be used to extract the maximum useful information for processing the large data, and it has gradually displayed its advantages in the related LIBS research field. However, the combination of LIBS and chemometrics has not been used for the quantitative analysis of plastics. Random forest based on variable importance (VIRF), the latest pattern recognition method based on classification trees or regression trees, has a good tolerance for noise and avoids the over-fitting phenomenon, and it has shown excellent performance in classification analysis. However, there are few reports on quantitative analysis using VIRF combined with LIBS. In this work, the combination of LIBS and random forest regression based on variable importance (VI-RFR) was used for the quantitative analysis of Pb, Cr, and Hg in PP. The spectral library consisted of 480 LIBS spectra from 6 types of plastics, with the spectra in the test set fixed and correlated versus the spectral data in the training set. Different pre-processing methods (normalization and mean centering) and variable importance were employed to improve the performance of VI-RFR for plastic analysis. To validate its performance for plastic analysis, VI-RFR was compared with random forest regression and partial least squares regression. VI-RFR exhibited the lowest root mean squared error and highest correlation coefficient, which indicated a better performance for the quantification of Pb, Hg and Cr in plastics.