Dataset of honey quality traits analysed with NIR

Honey quality and authenticity assessment require rapid and reliable analytical tools capable of supporting both laboratory and on-site applications. Near-infrared (NIR) spectroscopy represents a non-destructive and cost-effective approach; however, its performance depends on instrument characteristics and chemometric strategies. This study compared one benchtop and two portable NIR spectrometers for predicting key physicochemical parameters (moisture, electrical conductivity, glucose, fructose, reducing sugars, pH, hydroxymethylfurfural, and diastatic activity) and for discriminating botanical origin in 80 Italian honey samples. Spectral data were processed using multiple pre-processing techniques and algorithms (PLS, k-NN, Random Forest, SVM), with and without wavelength selection (siPLS and CARS-PLS), under cross-validation schemes. The benchtop device achieved the highest regression performance (R² up to 0.91 for glucose and electrical conductivity) and the most reliable botanical classification (balanced accuracy = 0.90). Portable instruments showed moderate predictive ability for bulk compositional parameters (R² up to 0.86 for glucose) but limited classification performance. Wavelength selection resulted in only marginal improvements. Parameters present at low concentrations, such as hydroxymethylfurfural and diastatic activity, were poorly predicted across all devices. These findings indicate that portable NIR systems are suitable for rapid screening of major compositional traits, whereas benchtop instruments remain more appropriate for laboratory-level quantification and robust botanical authentication.