Researchers in Bangladesh have developed a rapid, non-destructive method to detect honey adulteration using UV-Vis-NIR spectroscopy paired with machine learning. Their findings could protect consumers and support food quality enforcement.
• UV-Vis-NIR spectroscopy with machine learning reveals adulterants in Bangladeshi honey.
• The study analyzed five types of honey using non-destructive techniques.
• Common adulterants included corn syrup, glucose syrup, and caramel color.
• Model accuracy reached over 99% in detecting adulteration.
Introduction: Tackling a Growing Sweet Scam
In a country known for its rich natural resources, honey has become a prime target for food fraud. But now, a team of Bangladeshi scientists has developed a powerful method to detect adulteration in honey, without ever opening the jar. In a study published in LWT (1), researchers used ultraviolet-visible-near-infrared (UV-Vis-NIR) spectroscopy combined with machine learning (ML) to rapidly and non-destructively identify adulterants in five different types of honey commonly found in Bangladesh (1).
How They Did It: Spectroscopy Meets AI
Led by Al Noman, M.A., and a multi-institutional team including A.B. Nijhum, I. Hossain, M.S. Islam, I.M. Sifat, M.G. Aziz, and A. Rahman, the researchers collected samples of both pure and adulterated honey. Adulterants included corn syrup, glucose syrup, and caramel color—all commonly used to increase volume and sweetness while deceiving consumers.
Instead of relying on time-consuming and destructive chemical tests, the researchers turned to UV-Vis-NIR spectroscopy (1,2). This method captures the absorption spectra of honey samples across a wide range of wavelengths (200–900 nm), enabling detailed chemical profiling. When coupled with ML, the technique identifies patterns invisible to the human eye (1,2).
“The UV-Vis-NIR spectra of adulterated samples showed distinct deviations in the 200–900 nm range,” the authors noted, emphasizing that adulteration impacts both organic and inorganic content in honey. These spectral differences formed the basis for computational models capable of classifying and quantifying adulterants (1).
Machine Learning Powers Precision Detection
To analyze the spectral data, the team applied several preprocessing techniques—such as standard normal variate (SNV), Savitzky-Golay smoothing, and first and second derivatives—to reduce noise and enhance key features. The authors repor ttha "Spectral smoothing was applied using the Savitzky–Golay filter with an 11-point window and a second-order polynomial to reduce noise while preserving peak characteristics" (1). They further report that "baseline corrections were performed utilizing the Improved modified polynomial (IModPoly) algorithm implemented via the baseline removal package to correct baseline drift" (1). Finally, he intensities for spectra were normalized using a standard scaler.
They then trained a suite of ML algorithms, including Random Forest, support vector machines (SVM), and soft independent modeling of class analogy (SIMCA), achieving up to 100% classification accuracy (1).
The standout performer was the Random Forest (RF) model showed the highest accuracy, which achieved classification accuracies of (99–100 %), the study author’s reported.
Among the models, the authors specifically reported that "RF achieved the highest performance, with classification accuracies of 100 % for botanical origin, 99 % for adulteration presence, and 100 % for adulterant type" (1).
Implications for Food Safety and Market Integrity
This research offers a scalable and affordable solution for monitoring honey quality, critical in a country like Bangladesh, where local honey is both a staple and an export commodity. By using a non-destructive, fast method that requires no chemical reagents, the approach is not only environmentally friendly but also practical for routine field use (1).
“Combining spectroscopy with artificial intelligence holds enormous promise for food authentication,” the authors said. Their method could eventually be adapted to detect adulteration in other food products such as milk, oil, and fruit juices (1).
Conclusion: A Bright Future for Honest Honey
As consumers worldwide demand greater transparency in food sourcing, methods like this offer a high-tech safeguard against fraud. With accuracy levels nearing perfection and a completely non-destructive approach, UV-Vis-NIR spectroscopy paired with machine learning could soon become the gold standard for honey authentication, ensuring that what’s labeled “pure” really is (1).
References
(1) Al Noman, M. A.; Nijhum, A. B.; Hossain, I.; Islam, M. S.; Sifat, I. M.; Aziz, M. G.; Rahman, A. Non-Destructive Adulterants Detection in Various Honey Types in Bangladesh Using UV-Vis-NIR Spectroscopy Coupled with Machine Learning Algorithms. LWT 2025, 182, 118125. DOI:
(2) Gajdoš Kljusurić, J.; Knights, V.; Durmishi, B.; Rizani, S.; Jankuloska, V.; Velkovski, V.; Jurinjak Tušek, A.; Benković, M.; Valinger, D.; Jurina, T. Data Analyses and Chemometric Modeling for Rapid Quality Assessment of Enriched Honey. Chemosensors 2025, 13 (7), 246. DOI:
