Application of FTIR spectroscopy and chemometrics for the authentication of oils and margarines

Abstract: Rapid methods that can classify, discriminate, and authenticate food products at detection levels equal to or higher than those of currently used analytical methods is constantly needed. This study aimed to assess the combined utility of attenuated total reflectance-Fourier-transform infrared (ATR-FTIR) spectroscopy and machine learning (ML) techniques as a quick tool for identifying and classifying oils and margarines. Palm kernel oil (PKO), coconut oil (CCO), and njangsa seed oil (NSO) were used to formulate NSOCCO and NSOPKO margarines by blending. Fingerprints of the oil and the margarine derived in the spectra region 4000 - 600 cm^-1 were combined with chemometrics models including ML classifiers such as random forest (RF), logistic regression (LR), k-nearest number (KNN), decision tree (DT), support vector machine (SVM), and light gradient-boosting machine (LightGBM) models were used to classify, discriminate, and authenticate the oil and margarine samples. Principal component analysis (PCA) reduced the data dimension into two principal components, up to 99.4% and 98% for oil and margarine samples, respectively while enabling visualization. Margarine samples had 100% accuracy for all models except DT (50%), whiles the accuracy for oil classification was 97, 93, 83, 83, 53 and 50% for KNN, LR, SVM, LightGBM and DT, respectively. FTIR spectroscopy coupled with ML techniques offers a rapid method for the classification and authentication of oil and margarine samples.