Comparative evaluation of near-infrared, Fourier transform infrared, and nuclear magnetic resonance spectroscopy for λ-cyhalothrin screening in virgin olive oil

Pesticide residue control is an important component of regulatory surveillance in the olive oil market. This preliminary study evaluated near-infrared (NIR), Fourier transform infrared (FTIR), proton nuclear magnetic resonance (¹H NMR), and fluorine nuclear magnetic resonance (¹⁹F NMR) spectroscopy for the rapid screening of λ-cyhalothrin in virgin olive oil (VOO). A total of 220 samples covering 11 concentration levels from 0.1 to 1.4 mg/kg oil were prepared to represent a concentration range relevant to an EU compliance assessment example. Spectra obtained from the four analytical approaches were first examined to assess their sensitivity to λ-cyhalothrin-related spectral variation. No clearly distinguishable λ-cyhalothrin-specific signals were observed in either the ¹H or ¹⁹F NMR spectra under the applied experimental conditions, most likely because of the low pesticide concentrations and insufficient detection sensitivity. NIR and FTIR spectra were subsequently used to develop multiclass models for assigning samples to three study-defined maximum residue level-related screening groups, namely Below-MRL, Within-MRL, and Above-MRL, as well as regression models for quantitative prediction. The NIR-XGBoost model achieved an accuracy of 1.0 in both the calibration and independent prediction sets. For quantitative analysis, the NIR-SVM model provided the best predictive performance, with an Rp² of 0.909, an RMSEP of 0.113 mg/kg, and an RPD of 3.349. Predictions were relatively stable in clearly low-risk and high-risk concentration ranges, whereas greater uncertainty occurred near the study-defined screening thresholds of 0.37 and 1.11 mg/kg oil. FTIR models showed lower generalization ability, mainly because pesticide-related spectral variation overlapped with the dominant triglyceride absorptions. Overall, among the evaluated techniques, NIR spectroscopy combined with machine learning showed the greatest potential for rapid preliminary screening of λ-cyhalothrin in VOO before confirmatory chromatographic analysis, whereas conventional ¹H and ¹⁹F NMR showed insufficient sensitivity at the investigated residue levels.