Machine Learning-Based Classification of Microplastics Using Spectral Data for Environmental Monitoring

Accurate classification of microplastic particles is important for developing automated analytical methods. This study investigates machine-learning-based classification of five standard microplastic polymers—polyethylene (PE), polyethylene terephthalate (PET), polypropylene (PP), polystyrene (PS), and polyvinyl chloride (PVC)—using experimentally acquired laser-induced fluorescence (LIF) spectra. A measurement system combining 266 nm laser excitation, hydrodynamic focusing, and spectroscopic detection was used to collect 1,300 fluorescence spectra. Support Vector Machine (SVM) and Random Forest (RF) classifiers were evaluated using full-spectrum data containing 2,047 features and a restricted wavelength interval of 275–600 nm containing 553 features. In the stratified 80:20 hold-out evaluation, SVM and RF achieved accuracies of 98.46% and 98.08%, respectively. Both classifiers maintained 98.08% accuracy using the restricted spectral interval. Seven-fold Leave-One-Day-Out (LODO) cross-validation was additionally performed to assess acquisition-day variability. The LODO accuracies ranged from 96.46% for SVM without L2 normalization to 99.08% for L2-normalized RF, with the latter achieving 95.00% accuracy on its weakest acquisition day. These findings demonstrate the feasibility of machine-learning-based classification using experimentally acquired LIF spectra under controlled laboratory conditions. However, further validation using independently sourced and environmentally representative microplastic samples is necessary to establish broader applicability.