SELECT-OLS calibration models for robust fatty acids quantification in various vegetable oils using NIR spectroscopy: A unified approach across hydroxytyrosol supplementation and deep-frying conditions

Abstract

This study aims to develop a unified, reliable, and cost-effective model for quantifying fatty acids in edible oils to improve nutritional labeling and thermal-oxidation stability. Using near-infrared spectroscopy (NIRS) combined with multivariate calibration, reduced-spectrum regression models were created to identify key fatty acid markers in various oils, including extra virgin (EVOO), virgin (VOO), refined (ROO), and pomace olive oils, as well as sunflower oils. Oils were analyzed before and after deep frying, with or without hydroxytyrosol (HTyr) supplementation. Stepwise decorrelation of the variables (SELECT) improved regression models for seven fatty acids: myristic, palmitic, palmitoleic, stearic, oleic, linoleic, and lignoceric acids, achieving correlation coefficients (R) of 0.96–0.99. SELECT-ordinary least squares (OLS) regression provided highly predictive models, using up to 30 out of 700 wavelengths, with selected predictors ranging from 5 (for linoleic acid) to 30 (for stearic acid), showing adaptability to various spectral patterns. These models enable real-time fatty acid quantification in EVOO, VOO, ROO, and sunflower oils, under different deep-frying conditions (170–210 °C/3–6h), including both HTyr-supplemented and non-supplemented samples. NIRS with optimized variable selection simplifies analysis and reduces costs by replacing gas chromatography, enabling direct analysis without sample preparation. The models showed strong predictive performance, with high leave-one-out (LOO) explained variance (85.81 %–98.16 %), confirming NIRS as a feasible, rapid, non-destructive, and eco-friendly method for quantifying edible oils.