AI-enabled spectral classification of plastic resins from E-waste via laser-induced breakdown spectroscopy for advanced sorting applications

• Achieved 96 % (static) and 98 % (dynamic) LIBS classification accuracy. • Evaluated 8 ML algorithms on diverse samples across 6 resin classes. • C-247 and min–max normalization compared for LIBS spectral data. • NNMLP, SVM, and KNN showed good performance with consistent accuracy. In the framework of the circular economy (CE), efficient waste segregation is essential for sustainable recycling. Plastic waste from waste electrical and electronic equipment (WEEE) poses challenges due to complex resin structures and the presence of brominated flame retardants (BFRs). This study investigates the use of laser-induced breakdown spectroscopy (LIBS) combined with supervised machine learning (ML) for the classification of various e-waste plastics, including mixed resins and those containing BFRs. Although the primary analysis was conducted using static LIBS data, dynamic tests were also performed to simulate real-world sorting conditions. Among the classifiers, Support Vector Machine (SVM) and Neural Network Multilayer Perceptron (NNMLP) delivered the best results, reaching 92–94 % accuracy on test data and up to 96 % on unseen datasets. Furthermore, dynamic trials showed over 98 % accuracy, confirming the robustness of the approach. These findings highlight the potential of LIBS–ML systems for scalable, high-precision sorting, advancing industrial recycling strategies.