Answers are grounded in this paper and cite their sources.
Honey adulteration is a persistent global issue demanding rapid, accurate, and transparent analytical solutions. Machine learning (ML) coupled with spectroscopy offers a rapid solution for detecting honey adulteration, but the black box nature of advanced algorithms limits their adoption in regulatory quality control. Regulatory bodies require not only high accuracy but also transparent, chemically explainable decisions. This study establishes an interpretable ML framework for quantifying adulteration (Corn Syrup, Glucose Syrup, Caramel Color) in four distinct Bangladeshi honey varieties (wild, blackseed, mustard, rubber) using UV–Vis-NIR spectroscopy (200–900 nm), a large-scale dataset of 1960 samples was analyzed, featuring a high-precision adulteration range from 0% to 40% with 2.5% increments. A consensus feature selection strategy was employed, integrating Random Forest (RF), Recursive Feature Elimination with Cross-Validation (RFECV), and Shapley Additive Explanations (SHAP) to identify the most robust spectral markers. Five regression models - Light Gradient Boosting Machine (LightGBM), Categorical Boosting (CatBoost), Extreme Gradient Boosting (XGBoost), Random Forest (RF), K-Nearest Neighbors (KNN) were evaluated on this optimized feature set. LightGBM outperformed all other models, achieving an R 2 of 0.9375 and an RMSE of 0.0258. Critically, the consensus feature selection converged on eight key wavelengths (243, 373, 380, 382, 394, 576, 578, and 899 nm). SHAP analysis revealed that predictions were driven by the dilution of specific phenolic markers (UV region: 373–394 nm) and the enhancement of melanoidin absorption (Visible region: 576 nm), aligning ML decision-making with established chemical theory. By successfully decoding the spectral fingerprint of local honey varieties, this study demonstrates that “Explainable AI” can bridge the gap between high-performance computing and trustworthy food safety protocols. The proposed framework offers a replicable model for developing region-specific, low-cost sensors for honey authentication.
No takes yet. Share an insight, caveat, or question.
Noman et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: