Predicting oleogels properties using non-invasive spectroscopic techniques and machine learning

Oleogelators are considered food additives that require approval from regulatory authorities. Therefore, classifying these ingredients that may have characteristics (e.g., waxiness), cost and origin (e.g., animal or vegetable) is crucial to ensure consumer choice. In view of this, this study shows a non-invasive method for classification of oleogels based on several oleogelators, in addition to quantifying their concentration and their respective free fatty acid content and oil loss. To perform this quantification in a non-destructive, eco-friendly, portable, fast, and effective way, a colorimeter, a Raman spectrometer and 2 near-infrared spectroscopes with complementary ranges were used. Oleogels were prepared from sunflower and soybean oil, with different concentrations of 1 to 10 % (w/w) of beeswax, glycerol monostearate and ethylcellulose as oleogelators. After spectra pretreatment, Principal Component Analysis (PCA), classification and regression were performed. Random Forest (RF) models classified the samples based on which oil was utilized and the type of oleogelators with 100 % accuracy and their respective concentration with 94 % accuracy. The Partial Least Squares Regression (PLSR) for free fatty acid content and oil loss showed high performance, achieving residual predictive deviations (RPD) higher than 3 and range error ratios (RER) higher than 10 in the external validation set, indicating suitable predictive capacity and acceptability for quality control. The spectroscopic instruments, especially the colorimeter and NIR spectrometer, showed to be promising tools for monitoring these additives and predicting free fatty acid content and oil loss, ensuring the quality of these oleogels.