Detection of camellia oil adulteration with excitation-emission matrix fluorescence spectra and machine learning

Authors: WEI, CHAOJIE - China Agricultural University; WANG, WEI - China Agricultural University; JLABO, YANNA - China Agricultural University; Yoon, Seung; Ni, Xinzhi; WANG, XIAORONG - China Agricultural University; SONG, ZIWEI - China Agricultural University; HU, YATING - China Agricultural University

Submitted to: Journal of Food Composition and Analysis
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/24/2025
Publication Date: 3/10/2025
Citation: Wei, C., Wang, W., Jlabo, Y., Yoon, S.C., Ni, X., Wang, X., Song, Z., Hu, Y. 2025. Detection of camellia oil adulteration with excitation-emission matrix fluorescence spectra and machine learning. Journal of Food Composition and Analysis. https://doi.org/10.1007/s10895-025-04229-7.
DOI: https://doi.org/10.1007/s10895-025-04229-7

Interpretive Summary: Excitation-emission matrix fluorescence (EEMF) spectroscopy involves a 3D data scan that consists of excitation wavelengths, emission wavelengths, and fluorescence intensities. In this study, EEMF and machine learning were combined to identify the authenticity of pure camellia oil and predict the adulteration level when the pure camellia oil was adulterated by other cheap oils, such as palm oil, soybean oil, and maize oil. Parallel factor analysis and machine learning were used to analyze the spectral data in a factor space with five components. The correct classification rate of the support vector machine for identifying different types of edible oils was over 96.80%. The determination coefficient of validation of the partial least squares regression model for predicting adulteration levels was over 0.94 while the limit of detection was 2.59-4.23%. The study demonstrated that the adulteration of camellia oil could be detected accurately and its level could be also predicted with the EEMF spectroscopy combined with machine learning. The finding suggested the feasibility of a non-destructive and efficient tool for identifying the authenticity of camellia oil and measuring the adulteration level.

Technical Abstract: Camellia oil is a high-quality edible oil with significant nutritional and commercial value. However, it is facing increasing issues with authenticity due to adulteration practices. To ensure the health and benefit of consumers, there is a need for an accurate and non-destructive method to detect the types and concentration levels of adulteration. Excitation-emission matrix fluorescence (EEMF) spectra were obtained for pure edible oils, binary adulterated oils, and multi-component adulterated oils. A five-component parallel factor analysis (PARAFAC) model was utilized to roughly characterize the spectral properties of the edible oil samples and extract fluorophores. The results showed that the relative concentration of PARAFAC varied linearly with increasing adulteration concentration. Classification models were developed to qualitatively distinguish different types of edible oils based on EEMF spectra and relative concentration of PARAFAC. The support vector machine (SVM) model achieved an accuracy of over 96.80%. Furthermore, the partial least squares regression (PLSR) model, based on relative concentrations of PARAFAC, demonstrated good performance with a determination coefficient of validation (Rv2) > 0.94, a root mean square error of validation (RMSEv) < 6.03, a ratio of percentage deviation (RPD) > 4.12, and a limit of detection (LOD) of 2.59-4.23%. The results demonstrate that an accurate prediction of adulteration levels in pure Camellia oil is achievable. The combination of EEMF and machine learning has the potential to be a valuable tool for non-destructive, rapid qualitative and quantitative assessment of camellia oil adulteration.