Machine learning-assisted SERS detection of pyrethroid pesticides in edible fungi using a magnetic nanosensor

Pyrethroid pesticide residues pose a significant global public health challenge, particularly in complex edible fungus matrices where trace, structurally similar pesticides are difficult to distinguish and detect. To address this critical gap, a novel surface-enhanced Raman spectroscopy (SERS) platform is presented integrating "magnetic enrichment, covalent organic framework (COF)-mediated capture, silver shell enhancement, and machine learning (ML)-driven intelligent recognition". The core innovation lies in the tailored Fe₃O₄@COF@Ag nanocomposite architecture where the magnetic Fe₃O₄ core enables rapid sample separation, the COF intermediate layer provides specific molecular capture, and the Ag shell generates dense SERS "hotspots", synergistically enhanced by ML algorithms to enable precise discrimination of structurally analogous pesticides. The developed Fe₃O₄@COF@Ag substrate demonstrated exceptional performance, achieving detection limits as low as 5.21 µg/kg for deltamethrin (CF), 6.08 µg/kg for fenvalerate (FV), and 11.6 µg/kg for lambda-cyhalothrin (LCF). The method exhibited outstanding linearity (R² > 0.99) for quantitative analysis and anti-interference capability, with recoveries of 94.36-106.89% in real samples. By integrating ML algorithms, the platform achieved 100% classification accuracy for distinguishing structurally similar pyrethroids. Principal component analysis (PCA) and support vector machine (SVM) models effectively extracted discriminative spectral features, enabling precise identification even at trace concentrations. This work provides a rapid, sensitive, and field-deployable solution for pesticide monitoring, offering significant potential to enhance food safety and public health protection.