Interpretable Machine Learning in Food Analysis: Closed-Form Chemometric Calibration of Mid- and Near-Infrared Spectra Using Kolmogorov-Arnold Networks.
- 1Department of Mechatronics Engineering, Faculty of Mechanical Engineering, Yildiz Technical University, 34220 Istanbul, Türkiye.
- 2Department of Food Engineering, Faculty of Chemical and Metallurgical Engineering, Yildiz Technical University, 34220 Istanbul, Türkiye.
- 3Department of Nutrition and Dietetics, Faculty of Health Sciences, Biruni University, 34214 Istanbul, Türkiye.
Abstract
A Kolmogorov-Arnold network (KAN) learns a univariate function on every edge, so a trained network can be written as a closed-form equation, unlike a multilayer perceptron (MLP). A KAN was compared with five models, with selection nested within each withheld-lot or withheld-season fold, on tahini adulteration by attenuated total reflectance Fourier-transform infrared (ATR-FTIR) spectroscopy (1554 spectra, released here) and mango dry matter by near-infrared (NIR) spectroscopy (across-season test). No accuracy advantage was found: the KAN's coefficient of determination (R²) was comparable to a parameter-matched MLP's but never higher on average, and on withheld lots no difference between the leading models was significant. Most of the accuracy lost in converting the network to an equation was due to the conversion procedure; correcting it raised the median external R² of the tahini equation from 0.952 to 0.988. Repeated conversions of one network gave equations scoring -0.76 to 0.96, whereas an elementary edge basis makes the printed equation identical to the network up to rounding. With every lot serving as the blank, detection limits for samples resembling the calibration population were 8.9-21.6% adulterant, and blends at 4-12% were not quantified. The KAN provides an exact, inspectable screening calibration rather than higher accuracy.
Related
- Near-Infrared Spectroscopy in Food Analysis: Applications, Chemometric Strategies, and Technological Advances
- Towards Universal Calibration for Infrared Spectroscopy with Probabilistic Interference Removal
- Interpretable Machine Learning Disentangles Pore Confinement Across Multiple Zr-MOF CO2 Capture Targets
- Interpretable Machine Learning-Based QSPR Modeling for Predicting Hydrogen Adsorption Performance of Ester Oil Impregnating Agents
- Physics-inspired feature engineering and interpretable machine learning for thermoelectric properties of multicomponent materials
BibTeX
@article{batayhi2026interpretable,
title = {{Interpretable Machine Learning in Food Analysis: Closed-Form Chemometric Calibration of Mid- and Near-Infrared Spectra Using Kolmogorov-Arnold Networks.}},
author = {Abdulhamid Batayhi and Muhammed Özgölet and Hasan Cankurt and Muhammet Arici},
journal = {Foods (Basel, Switzerland)},
year = {2026},
month = oct,
doi = {10.3390/foods15203639},
url = {https://europepmc.org/article/MED/42940276},
note = {CC BY},
}
RIS
TY - JOUR
TI - Interpretable Machine Learning in Food Analysis: Closed-Form Chemometric Calibration of Mid- and Near-Infrared Spectra Using Kolmogorov-Arnold Networks.
AU - Abdulhamid Batayhi
AU - Muhammed Özgölet
AU - Hasan Cankurt
AU - Muhammet Arici
PY - 2026
DA - 2026/10/13
JO - Foods (Basel, Switzerland)
DO - 10.3390/foods15203639
UR - https://europepmc.org/article/MED/42940276
AB - A Kolmogorov-Arnold network (KAN) learns a univariate function on every edge, so a trained network can be written as a closed-form equation, unlike a multilayer perceptron (MLP). A KAN was compared with five models, with selection nested within each withheld-lot or withheld-season fold, on tahini adulteration by attenuated total reflectance Fourier-transform infrared (ATR-FTIR) spectroscopy (1554 spectra, released here) and mango dry matter by near-infrared (NIR) spectroscopy (across-season test). No accuracy advantage was found: the KAN's coefficient of determination (R²) was comparable to a parameter-matched MLP's but never higher on average, and on withheld lots no difference between the leading models was significant. Most of the accuracy lost in converting the network to an equation was due to the conversion procedure; correcting it raised the median external R² of the tahini equation from 0.952 to 0.988. Repeated conversions of one network gave equations scoring -0.76 to 0.96, whereas an elementary edge basis makes the printed equation identical to the network up to rounding. With every lot serving as the blank, detection limits for samples resembling the calibration population were 8.9-21.6% adulterant, and blends at 4-12% were not quantified. The KAN provides an exact, inspectable screening calibration rather than higher accuracy.
SN - 2304-8158
ER -