Application of ¹³C NMR and Untargeted Multivariate Analysis for Classification of Virgin Coconut Oil

Virgin coconut oil (VCO) is produced from fresh coconut meat without the use of chemicals and high heat. To ensure quality, it is important to be able to differentiate fresh quality VCO from old VCO, refined, bleached and deodorized coconut oil (RBDCO), and VCO which is adulterated with RBDCO. Differentiating these types of samples is a challenge because of their chemical similarity. This study investigated the use of ¹³C NMR and multivariate analysis to differentiate these samples. The methodology used the standard ¹³C NMR pulse sequence with broadband ¹H decoupling using CDCl₃ as solvent and dioxane as quantitative internal standard (IS). After pre-processing the spectra (alignment, bucketing/binning, normalization with respect to dioxane IS peak), untargeted multivariate analysis – unsupervised and supervised – were done on the bins of the ¹³C peaks. The linear unsupervised methods were able to differentiate fresh VCO (as control, 51 samples) from RBDCO (19 samples), adulterated VCO (12 samples), and VCO which was over 2 years old (5 samples). These results were verified by performing supervised binary linear classifiers using a one versus the rest strategy. Using AUC-ROC curves, the supervised dataset gave better than 95% accuracy in differentiating fresh VCO versus RBDCO, and better than 90% accuracy in differentiating fresh VCO versus adulterated VCO and old VCO using cross-validation and training holdout. Using cross-validation, this method was also able to differentiate with up to 73% accuracy among the three processes which were used to make the VCO samples (fermentation, centrifuge, and expeller). We conclude that linear techniques are adequate to classify fresh VCO from RBDCO, adulterated VCO, and old VCO.