Classification of oils and margarines by FTIR spectroscopy in tandem with machine learning.

[Display omitted] • ML approaches were extended to food identification and adulteration detection. • FTIR in combination with ML offered a non-destructive path for classifying selected oils and margarines. • PCA efficiently separated oil and margarine samples. • FTIR-ML method could detect 1% adulteration in oil samples. • KNN had the highest accuracy and R2 in both pure and adulterated samples. This study assessed the combined utility of ATR-FTIR spectroscopy and machine learning (ML) techniques for identifying and classifying pure njangsa seed oil (NSO), palm kernel oil (PKO), coconut oil (CCO), njangsa seed oil-palm kernel oil (NSOPKO) and njangsa seed oil-coconut oil (NSOCCO) margarine. Additionally, it quantified the degree of adulteration in each oil and margarine using ML regression models and sunflower oil and canola-flaxseed oil margarine as adulterants. Fingerprints of the oils and the margarines derived in the spectra region 4000–600 cm−1 were combined with ML models. The first two principal components explained 99.4% and 98% of the variance of pure oils and margarines and 90.1, 88.3, 88, 97.3 and 98.3% of adulterated PKO, NSO, CCO, NSOCCO and NSOPKO, respectively while enabling visualization. Pure margarines were classified accurately (100%) in all models. KNN was most effective in classifying pure oil at 97% followed by LR (93%), SVM (83%), LightGBM (53%) and DT (50%). The R2 obtained from all the models for adulterated PKO, NSO, CCO, NSOPKO and NSOCCO ranged from 59–99%, 55–99%, 45–94%, 69–98% and 59–94%, respectively. SVM and DT underperformed, while KNN was the best model. [ABSTRACT FROM AUTHOR]