Near-Infrared Spectroscopy Shows Potential for Detecting Mineral Oil Contamination in Edible Peanut Oil

Recently, a team of researchers from Jiangsu University and Jimei University in China, examined how to improve detection of oil-based contaminants in peanut oil. This study, which was published in Food Chemistry, showcased how Fourier transform near-infrared (FT-NIR) spectroscopy, when integrated with advanced chemometric models, can accurately identify and quantify five common contaminants in peanut oil (1).

How do mineral oil contaminants enter edible oils such as peanut oil?

Mineral oil contaminants are abundant because of industrial processes and transportation systems. Industries such as manufacturing, truck transportation, and textiles are particularly notable for the contaminants that they emit through normal practices, even though the regulations in place help prevent most of these contaminants from getting into the environment or food supply. However, despite preventative efforts, mineral oil contaminants can still enter the food chain through improper handling or accidental leaks (1).

Edible oils, such as vegetable and peanut oil, not only help food taste better, but they also contain important nutrients (2). Numerous studies have examined how spectroscopy can play a role in improving the overall quality of edible oils (2,3). This study continues this exploration, focusing primarily on five mineral oil contaminants: diesel, kerosene, lubricating oil, engine oil, and white mineral oil

Peanut oil is often used in the fast-food industry and was chosen as the matrix for the analysis in this study because of its stable properties and extensive use (1).

What was the experimental procedure?

As part of the experimental procedure, the researchers prepared 450 contaminated samples using nine concentration levels, ranging from 0.5 mg/kg to 30 mg/kg (1). These were mixed using a multi-step homogenization process involving manual shaking, mechanical stirring, and ultrasonic treatment to ensure uniform dispersion. Using a FT-NIR spectrometer from Thermo Fisher Scientific, the researchers collected the spectral data in the 4000–10,000 cm⁻¹ range, with each sample scanned three times to ensure reproducibility (1).

Meanwhile, out of the models used for qualitative analysis, partial least squares discriminant analysis (PLS-DA) was the standout, achieving 100% accuracy in distinguishing normal (pure) from contaminated samples across all five contaminant types (1). Even in cases involving engine and lubricating oils, which typically contain additives that can complicate spectral interpretation, PLS-DA maintained its accuracy. When tested against other classification models, which included k-nearest neighbors (KNN), support vector machines (SVM), and backpropagation neural networks (BPNN), PLS-DA and SVM consistently outperformed the rest in predicting specific contaminant types in new samples (1).

For quantitative analysis, the research team used two regression techniques. These regression techniques were support vector regression (SVR) and partial least squares regression (PLSR). SVR emerged as the more powerful tool for analyzing nonlinear relationships in the spectral data (1). It achieved correlation coefficients (R_p) greater than 0.98 for diesel, white mineral oil, and lubricating oil. PLSR also performed well, particularly for kerosene (R_p = 0.9335) and engine oil (R_p = 0.9270) (1). These findings indicate that NIR spectroscopy, when paired with the appropriate modeling techniques, is a reliable tool for both detecting and measuring trace levels of mineral oil contamination in food products.

What challenges did the research team encounter in the study?

While conducting the study, the research team acknowledged that they ran into issues at lower contamination levels. At the lowest tested concentration of 0.5 mg/kg, prediction accuracy decreased slightly, especially for kerosene and engine oil (1). Because of this, the authors recommended that future research should explore more sophisticated algorithms, such as ensemble learning or deep learning, to improve detection capabilities at these lower thresholds (1). They also suggested expanding this approach to other edible oils such as soybean, sunflower, and olive oil to develop broader food safety applications (1).

By integrating FT-NIR spectroscopy with machine learning models, the researchers see this method as a potential solution for various stages of the supply chain, including production and processing to packaging and retail (1). Given the increasing global concern about food adulteration and contamination, the study presents a new method that can conduct high-throughput, non-invasive food analysis.

References

1. Deng, J.; Jiang, H.; Chen, Q. Qualitative and Quantitative Analysis of Mineral Oil Pollution in Peanut Oil by Fourier Transform Near-infrared Spectroscopy. Food Chem. 2025, 469, 142590. DOI: 10.1016/j.foodchem.2024.142590
2. Wetzel, W. Conducting Smarter Vegetable Oil Analysis Using NMR and Chemometrics. Spectroscopy. Available at: https://www.spectroscopyonline.com/view/conducting-smarter-vegetable-oil-analysis-using-nmr-and-chemometrics (accessed 2025-07-08).
3. Wetzel, W. AI-Powered Detection System Identifies Petroleum Contamination in Edible Oils. Spectroscopy. Available at: https://www.spectroscopyonline.com/view/ai-powered-detection-system-identifies-petroleum-contamination-in-edible-oils (accessed 2025-07-08).