Integration of UV-vis spectroscopy and machine learning for identification of recycled polyethylene terephthalate

The circular economy has driven renewed interest in polyethylene terephthalate (PET) recycling, including recycled PET for food packaging, with a focus on reducing production costs and meeting social development goals. This study presents a novel approach to classifying virgin and recycled PET using an economically viable ultraviolet-visible spectroscopy combined with machine learning algorithms. The results show that Baseline Removal (RMBL) is the optimal preprocessing method for binary classification, and Principal Component Analysis (PCA) combined with Random Forest (RF) is the most effective binary classification model to distinguishes between virgin and 100 % recycled PET. To further improve the capability to detect samples of low recycled content, a multi-classification model was then developed. This approach enables the detection of recycled PET content as low as 10 %, providing a quantitative possibility for the classification. This study demonstrates the effectiveness of this approach and has significant implications for sustainable recycling practices.