Abstract
The use of non-decontaminated recycled poly(ethylene terephthalate) (PET) in food packages arouses consumer safety concerns, and thus is a major obstacle hindering PET bottle-to-bottle recycling in many developing regions. Herein, machine learning (ML) algorithms were employed for the discrimination of 127 batches of virgin PET and recycled PET (rPET) samples based on 1247 volatile organic compounds (VOCs) tentatively identified by headspace solid-phase microextraction comprehensive two-dimensional gas chromatography quadrupole-time-of-flight mass spectrometry. 100% prediction accuracy was achieved for PET discrimination using random forest (RF) and support vector machine (SVM) algorithms. The features of VOCs bearing high variable contributions to the RF prediction performance characterized by mean decrease Gini and variable importance were summarized as high occurrence rate, dominant appearance and distinct instrument response. Further, RF and SVM were employed for PET discrimination using the simplified input datasets composed of 62 VOCs with the highest contributions to the RF prediction performance derived by the AUCRF algorithm, by which over 99% prediction accuracy was achieved. Our results demonstrated ML algorithms were reliable and powerful to address PET adulteration and were beneficial to boost food-contact applications of rPET bottles.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have