Automated data analysis and machine learning-driven assessment of environmental microplastics by TGA-FTIR

Microplastics persist as a ubiquitous environmental contaminant, and techniques which can rapidly quantify and identify microplastic content in the environment are of importance to properly assess their environmental and health impact. Many of the popular techniques to identify microplastics broadly fall under either vibrational spectroscopy, or thermoanalytical techniques; thermogravimetric analysis (TGA) coupled with Fourier transform infrared spectroscopy (FTIR) sits at the intersection. However, it remains relatively underutilized in the microplastics space, despite the substantial data generated by this technique. Each thermogram is also associated with ~200 FTIR spectra which can be rapidly assessed with targeted automated data analysis. This work explores the development of data analysis routines specialized in identifying plastic components from TGA-FTIR data. A dedicated spectral library and matching algorithm were developed to identify polymers from their gas-phase FTIR spectra . This was extended to the exploration of machine learning (ML) classification techniques such as support vector machine (SVM), random forest (RF), and multilayer perceptron (MLP). ML techniques were found to provide rapid, unambiguous identification in comparison a custom spectral matching algorithm. With either given approach, the identities can be correlated with mass-loss in the thermogram, combining a qualitative information with pseudo-quantitative analysis allowing for the rapid assessment of plastic content in a given sample.