Machine learning-assisted spectral optimization for improved PARAFAC modeling in fluorescence-based determination of carbamate pesticides

Abstract

The extensive use of carbamate pesticides such as aminocarb and carbaryl has raised increasing concerns due to their persistence and potential impact on environmental and food safety. Their inconsistent regulation and occurrence at trace levels in complex matrices highlight the need for sensitive and reliable analytical method­ ologies for their determination. Here, a fluorescence-based chemometric strategy integrating PARAllel FACtor analysis (PARAFAC) analysis with a machine learning (ML)-assisted workflow is proposed, where ML is used to guide spectral variable selection, leading to improved PARAFAC performance and enhanced second-order cali­ bration in complex matrices. Excitation–emission matrices (EEMs) were decomposed using PARAFAC, while ML was employed to unfold the data, select optimal linear models, and identify informative spectral regions. This approach enabled selective resolution and accurate quantification of both analytes in tap water and orange juice, even in the presence of matrix interferences. The method showed excellent predictive performance, with co­ efficients of determination (R2) higher than 0.99 and 0.98 for tap water and orange juice, respectively, REP values below 10% and 15%, and recovery values ranged from 94 to 107%. ML-assisted spectral optimization significantly improved PARAFAC performance in complex matrices in reducing interferences. Limits of detection of 7 nM for carbaryl and 72 nM for aminocarb were achieved, comparable to conventional methods. Similar results were obtained against an HPLC–FLD reference method, while improved sustainability was demonstrated by a higher AGREE score (0.77 vs 0.64). The proposed ML-guided PARAFAC strategy provides a robust, accurate, and environmentally friendly alternative for carbamate simultaneous determination in complex matrices.