Predictive framework for degradation-aware plastic classification using spectroscopy and machine learning for proper recycling

Plastics degradation strongly affects recyclability, yet conventional sorting classifies materials only by polymer type, overlooking degradation history and contributing to downcycling post-consumer recycled materials. Here, we present a nondestructive framework for assessing plastic age using spectroscopy, machine learning, and additive-based chemical design. Mildly aged commercial plastics showed no measurable changes by IR spectroscopy or thermogravimetric analysis, highlighting the limitations of conventional characterization methods. Separately, a Raman-based deep learning workflow developed using a publicly available weathered plastics dataset enabled proof-of-concept classification of stable and weathered spectra. Figure2CNN achieved 90.16% overall accuracy and, unlike ResNet1D, avoided complete majority-class collapse by correctly identifying spectra from both classes. An interactive inference dashboard was developed, and triazole-based UV absorbers are proposed as “age barcode” additives.