Classification of real plastic waste using machine learning-assisted laser-induced breakdown spectroscopy

Plastic type classification using laser-induced breakdown spectroscopy (LIBS) was evaluated using actual waste samples under non-optimized experimental conditions designed to approximate laser-sample variability encountered in industrial operation. The study addresses challenges in plastic sorting crucial for enhancing recycling efficiency and mitigating \:co₂ emissions and environmental harm from plastic waste overflow. Held-out testing using physically separate real-waste specimens was performed using data from standard materials for modeling and data from real waste for the test. The target plastic types were polypropylene, polyethylene terephthalate, high-density polyethylene, and low-density polyethylene, which are major types of plastic waste. Plastic sorting was assisted by two machine-learning techniques: principal component analysis and k-nearest neighbors. Twenty-three pre-processes and seven variable selection methods were examined to enhance accuracy. The synergistic impact of preprocessing and variable selection was evident, elevating accuracy from 0.617 to 1.00. This study demonstrates the potential of machine learning-assisted LIBS for real plastic waste classification, emphasizing the importance of relevant wavelength selection.