From Lab to Field: A Pilot Study Using Portable NIR and ML for Honey Botanical Traceability

Authentication of mono-floral honeys is still an open challenge in the agribusiness sector, especially in light of increasing market fraud and the limitations of traditional laboratory analytical techniques. This pilot study evaluated the applicability of two handheld near-infrared (NIR) sensors - SpectraPod and SCiO - for the classification of honey samples according to their botanical origin. The approach combines miniaturized optical sensing with chemometric strategies and supervised machine learning (ML) models. A dataset of six monofloral honeys was analyzed via diffuse reflectance NIR spectroscopy. Tailored preprocessing pipelines were developed for each sensor, taking into account differences in spectral resolution and detector response. Classification was performed using different ML models: Support Vector Machines (SVM), Random Forest (RF) and PLS-DA. The RF model achieved 90% accuracy, while PLS-DA provided chemically consistent results aligned with known vibrational modes of key components such as glucose and fructose. These preliminary results highlight the potential of combining miniaturized NIR sensing with machine learning for rapid, non-destructive infield honey authentication, while indicating the need for further validation on larger datasets.