Machine Learning-Assisted 3D Envelope-Enhanced SERS Platform for Detection and Classification of Multiple Coexisting Pollutants

Antibiotics in the environment readily adsorb onto small surface-area plastic surfaces, generating coexisting pollutants that pose greater risks than individual pollutants. Here, we developed a three-dimensional envelope-enhanced SERS platform (3D EES) strategy for the detection and quantification of coexisting pollutants by integrating surface-enhanced Raman spectroscopy (SERS) with machine learning (ML). The 3D EES strategy was established using three-phase interface self-assembly to organize Au@Ag plasmonic nanoparticles into a hydrophobic array substrate, generating dense and uniform plasmonic gap hot spots. The coconcentration of coexisting pollutants and plasmonic nanoparticles on the hydrophobic array substrate generated additional hotspots, further amplifying the SERS signal. This strategy enabled the detection of antibiotics adsorbed onto polystyrene (PS) at concentrations as low as 250 ng/L. A comprehensive SERS data set of coexisting pollutants was constructed, and machine-learning models (SNE-SVM and RF-SVM) achieved 98.07% accuracy and 96.44% Jaccard similarity, allowing reliable classification of antibiotic species adsorbed on plastic and accurate prediction of their adsorption levels. This method provides a robust tool for identifying coexisting pollutants in complex environments and offers new opportunities for tracing pollutant sources and understanding their environmental migration and transformation.