A semi-quantitative one-class partial least squares model for detecting honey adulteration using TD NMR spectroscopy

Identifier:  imarina:9453707

Handle: https://hdl.handle.net/20.500.11797/imarina9453707

Authors:  Rovira, G; Botelho, CSWM; de Oliveira, LL; Andrade, MVD; Santos, PM; Sena, MM; de Souza, SVC; Callao, MP; Ruisanchez, I

Abstract:
Honey adulteration with inverted sugar syrup is a common fraud that challenges current detection methods, which are often time-consuming, destructive, or lack clear decision frameworks. This study presents a rapid, nondestructive screening method using Time-Domain Nuclear Magnetic Resonance (TD NMR) combined with oneclass partial least squares (OCPLS) classifier. The novelty of this approach lies in defining two decision thresholds, creating three classification zones (non-adulterated, adulterated, and uncertainty), rather than relying on a single limit as in previous studies. Semi-quantitative performance parameters (decision limit, CC alpha; detection capability, CC beta; and unreliability region, UR) were determined from Performance Characteristic Curves (PCC), allowing the evaluation of model sensitivity to varying adulteration levels. Decision limits of 5 % and 2 % adulteration were established for eucalyptus and wild honey, respectively. This strategy eliminated misclassifications, ensuring 100 % reliability outside the uncertainty region, with inconclusive samples referred for confirmatory analysis. The proposed methodology improves upon existing methods by introducing objective, concentration-based decision criteria and a structured classification framework, offering a robust tool for reliable and practical honey authentication.