Výskumný ústav potravinársky

Ali, I. – Abdullah, R. – Zahid, T. – Kaleem, A. – Iqtedar, M. – Iqbal, I.
Integrating physicochemical analysis and artificial intelligence prediction to monitor degradation of edible oils during repeated frying and storage
Journal of Food and Nutrition Research, 65, 2026, č. 3, s. 216-229

Roheena Abdullah, Department of Biotechnology, Lahore College for Women University, Jail Road, Lahore 54000, Pakistan. E-mail: roheena.abdullah@lcwu.edu.pk
Imran Ali, The Center for Professional Education of Teachers-Youch, Future Education Institute, Daerwen Road No. 188, 200120 Shanghai, China; Institute of Molecular Biology and Biotechnology, University of Lahore, Raiwand Road, Lahore 54792, Pakistan; Institute of Biochemistry, University of Balochistan, Sariab Road, Quetta 87300, Pakistan. E-mail: imranalisheik@gmail.com

Original article
Received 13 April 2026; 1st revised 12 May 2026; 2nd revised 18 June 2026; accepted 30 June 2026; published online 24 July 2026

DOI: https://doi.org/10.64122/SZTC7474

This article contains supplementary data.

Súhrn: This study evaluated physicochemical changes in edible oils during repeated frying and storage using a hybrid analytical framework integrating experimental analysis with artificial intelligence (AI)-based predictive modelling. Eleven edible oils were analysed for peroxide value, iodine value, acid value, free fatty acids, viscosity, refractive index, moisture content, and specific gravity using standard methods. Repeated frying significantly increased peroxide value, acid value, free fatty acids, and viscosity, indicating oxidative and hydrolytic degradation, while iodine value decreased due to thermal oxidation of unsaturated fatty acids. Storage conditions affected oil stability; oils in transparent plastic bottles under light showed greater oxidative deterioration than those in dark glass containers. To predict oil degradation, random forest regression was benchmarked against multiple linear regression, support vector regression, and gradient boosting. Random forest achieved the best predictive performance (coefficient of determination R² = 0.92, root mean square error RMSE = 0.52), with peroxide value combined with frying cycle number (34 %), acid value with free fatty acids (26 %), and viscosity with specific gravity (18 %) as primary predictor groups. These findings demonstrate the practical utility of AI-assisted predictive monitoring for rapid edible oil quality assessment.

Kľúčové slová: physicochemical properties; peroxide value; frying stability; edible oils; artificial intelligence; random forest model

Na stiahnutie:
  jfnr-2026-3-pp216-229-ali.pdf (PDF, 693.12 Kb, 162x)
  jfnr-2026-3-pp216-229-supplement.pdf (PDF, 223.91 Kb, 183x)