Detection of Honey Adulteration Using Thermorheological and Spectroscopic Analyses: Independent Evaluation With...

Honey adulteration is a prevalent economic fraud that demands robust and reliable detection methods. In this study, a proof-of-concept was developed on a comparative study of oscillatory thermorheology (20°C-80°C) and spectroscopic techniques, including UV-Vis and fluorescence spectroscopy, for the detection of adulteration with glucose, invert sugar, and maltose syrups. A total of 86 samples were analyzed, comprising authentic blossom and honeydew honeys, as well as samples adulterated at concentrations ranging from 5% to 50%. Thermorheological and spectroscopic datasets were analyzed independently to assess their individual discriminative power. For each analytical approach, classification performance was evaluated using linear latent-variable models (PLS-DA, OPLS-DA) and gradient-boosting machine learning techniques (LightGBM, XGBoost), enabling a systematic comparison between classical chemometric and nonlinear machine-learning methods. The results revealed clear performance differences between the analytical approaches and modeling strategies. The results demonstrated that while spectroscopic models achieved high sensitivity (>0.96), they frequently failed to correctly identify authentic samples, resulting in notably low specificity (0.17-0.42). In contrast, thermorheological parameters more effectively captured the structural and physicochemical alterations induced by the addition of glucose, invert sugar, and maltose syrups. Notably, loss modulus data modeled using the LightGBM algorithm achieved the most balanced classification performance, reaching 0.75 specificity and 92.67% external validation accuracy while maintaining high sensitivity. These findings demonstrate that thermorheological analysis, particularly when evaluated using advanced non-linear gradient boosting models, offers superior discriminative capability for honey authentication compared to spectroscopic analysis alone.