INFORM May 2026

38 INFORM MAY 2026, VOL. 37, NO. 5

FTIR SPECTROSCOPY COUPLED WITH MACHINE LEARNING FOR ASSESSMENT

ELECTRONIC NOSE AND MACHINE LEARNING FOR RAPID, SUSTAINABLE DETECTION OF AVOCADO OIL ADULTERATION Mayorga-Martínez, A.A., et al. , Journal of Agricultural and Food Research , 27, 102775, 2026. Avocado oil, regarded as a health promoting food product due to its bioactive compounds, faces a persistent problem of adulteration. Existing methods for assessing the authenticity of edible oils are often complex, expensive, and time consuming. Furthermore, they are environmentally burdensome due to chemical wastes. This study used gas chromatography (GC) analysis, a traditional method, to evaluate 25 commercial samples for compliance with the fatty acid (FA) and phytosterol (PS) standards by Codex Alimentarius. Results showed that only 40% of the samples fully complied with the CA, confirming widespread non-authenticity, particularly among the private label brands. A low-cost electronic nose (e-nose) was employed to develop machine learning (ML) models for rapid authentication, using the GC data as reference. A principal component analysis (PCA) showed clear separation among samples based on fault levels. Significant correlations (R) were found between the e-nose sensor responses and key FAs and PSs. Classification models using artificial neural networks (ANN) achieved overall accuracies >95%. The developed ANN regression models for individual compounds, including oleic acid and β -sitosterol, showed also high accuracy (R = 0.92-0.98). Both classification and regression models showed no signs of overfitting. Mean squared error values during training were consistently lower than those obtained during testing. These approaches represent a cost effective, rapid, reliable, and scalable alternative for routine authenticity screening. It offers promising applications for the food industry, particularly for private-label retailers seeking to prevent food fraud and ensure product integrity.

In sesame oil, C18:1 and C18:2 were dominant, but adulteration elevated C16:0 levels, indicating the addition of oils richer in saturated fats. Principal component analysis (PCA) and hierarchical cluster analysis (HCA) effectively differentiated pure and adulterated samples, confirming the consistency of spectral changes across oil types. The combined use of FTIR, dielectric analysis, and explainable machine learning models offers a fast, reagent-free, and interpretable solution for edible oil quality assessment. MACHINE LEARNING ASSISTED INTELLIGENT IDENTIFICATION STRATEGY FOR ADULTERATION PEANUT OIL BASED ON FATTY ACID GC FINGERPRINT Su, G., et al. , Food Chemistry: X , 34,103686, 2026. This study established an intelligent identification method for adulterated peanut oil using machine learning. A total of 32 pure peanut oil samples and 126 adulterated samples containing 5%–30% soybean, palm, cottonseed, or sunflower oil were prepared. The fatty acid profiles of pure and adulterated oils were highly similar, making them difficult to distinguish effectively using principal component analysis or heatmaps. By applying four supervised machine learning algorithms (support vector machine (SVM), random forest, partial least squares, and decision tree), the classification accuracy is improved significantly. The SVM model performed best, achieving 98.18%–100% accuracy for both single and mixed adulteration samples. SHAP analysis identified lignoceric acid (C24:0) as the key adulteration marker. The regression model yielded R 2 values of 0.9153 and 0.7254 on the training and test sets, respectively. This method provides an accurate, interpretable approach for identifying peanut oil adulteration.

OF DIELECTRIC AND PHYSICOCHEMICAL

PROPERTIES OF PURE AND ADULTERATED COCONUT AND SESAME OILS Varma, S.O., et al. , Food Analytical Methods , 19, 116, 2026. Adulteration of edible oils poses significant health risks and economic concerns, highlighting the need for rapid, reliable, and non-destructive detection techniques. This study integrates Fourier Transform Infrared (FTIR) spectroscopy, microwave-based electric parameter measurements, and gas chromatography (GC) with multivariate chemometric methods to detect and quantify palm oil adulteration in coconut and sesame oils. Quality parameters, namely dielectric constant (DC), dielectric loss (DL), iodine value (IV), and refractive index (RF), were predicted using partial least squares regression (PLSR). The models achieved high validation accuracy for DC, DL, and RF ( R 2 >0.87). Key FTIR spectral bands at 1654.9 cm ⁻ 1 and 1099.4 cm ⁻ 1 , associated with fatty acid composition, were identified as important markers to differentiate types of oils. SHAP (SHapley Additive exPlanations) and Variable Importance in Projection (VIP) analyses identified FTIR bands associated with ester C=O and unsaturated C=C vibrations as the most influential features. These bands reflect changes in fatty acid saturation and ester composition between pure and adulterated oils, providing chemically interpretable markers for reliable discrimination of adulteration in edible oil. However, all parameters were validated using PLSR with GC spectra as the input, showing high accuracy as R 2 > 0.92. In pure coconut oil, GC revealed lower palmitic acid (C16:0), which increased with adulteration, along with higher oleic (C18:1) and linoleic (C18:2) acids.

C

M

Y

CM

MY

CY

CMY

K