Discrimination of the pollution grade of metal elements in atmospherically deposited particulate matter via laser-induced breakdown spectroscopy combined with machine learning method

Abstract

The harm of air pollutants to human body is obvious, which may cause various cancers, acute poisoning and even affect mental health. Therefore, the measurement of harmful elements and identification their pollution degree in atmospheric particulate matter are of great significance. In this research, taking the pollution grade of metal elements divided by the values of enrichment factor (EF) as the research target, we established the novelty method of LIBS combined with machine learning methods for fast and accurate discrimination of the pollution grade of metal elements in atmospherically deposited particulate matter. Firstly, the LIBS spectra of 17 atmospherically deposited particulate matter samples were measured, and the main metal elements were identified by national institute of standards and technology (NIST) database. Next, the concentration of main metal elements (Cu, Zn, Pb, Cr and V) was obtained by inductively coupled plasma mass spectrometry (ICP-MS), and then the pollution grade of metal elements was classified according to the EF value. Then the PCA method was applied, and it was found that PCA could not distinguish the samples with different pollution grades of metal elements. In order to improve the accuracy of discriminant analysis, the RF discriminant model was established for discriminant analysis of the pollution grade of metal elements, with emphasis on the influence of different pretreatment methods on the accuracy of RF discriminant model. Finally, under the optimal conditions of optimal spectral preprocessing method and model parameters, the RF and PLS-DA discriminant models were constructed for the discrimination of the pollution grade of metal elements. The results showed that compared with PCA and PLS-DA discriminant models, RF discriminant model had high accuracy discriminant results. For Zn discriminant analysis (Grade-3 and Grade-4) of prediction set, the accuracy, sensitivity and specificity were 0.9200, 0.9000, 0.9333, respectively. For Cr discriminant analysis (Grade-2 and Grade-3), the accuracy, sensitivity and specificity were 0.9200, 1.0000 and 0.9000, respectively. The results infer that LIBS technology combined with RF can be adopted to accurate and quick distinguishment the atmospherically deposited particulate matter samples with types of the pollution grade of metal elements, which helps to provide early warning for urban air pollution prevention and control.