Random Forest Regression assisted Raman spectroscopy for authenticating the purity of virgin coconut oil

Virgin coconut oil (VCO) adulterated with refined, bleached, and deodorised coconut oil (CO-RBD) is a major issue in the edible oil markets. Raman spectroscopy is a popular non-destructive, and selective technology for rapid detection of food adulterants. In this study, an ensemble-based machine learning method, random forest regression (RFR) was used to quantify the level of adulteration. Initially, Raman spectral features able to distinguish VCO and CO-RBD were extracted using principal component analysis (PCA). PCA revealed that the Raman spectral features related to =C−H bending, C C stretching of the cis RHC CHR group and C O stretching of the RC OOR group were important for classification. The RFR with PCA extracted features performed significantly better (mean absolute error (MAE) = 0.4043 and R 2 = 0.9993) compared to Partial Least Square regression (MAE = 5.7254, R 2 = 0.9391), Ridge regression (MAE = 5.3151, R 2 = 0.9396), LASSO regression (MAE = 5.8010, R 2 = 0.9315) or RFR only (MAE = 3.0281, R 2 = 0.9655). The limit of detection of the adulterant was as low as 1 %. Raman spectra subjected to RFR combined with PCA features therefore enables detection of low levels of CO-RBD in VCO where traditional indicators are limited. • VCO adulterated with CO-RBD compromises VCO quality. • Raman spectroscopy was used to obtain spectral signatures of VCO and CO-RBD. • Regression models including Random Forest Regression (RFR) was used to quantify the adulterant. • Principal Component Analysis (PCA) was used for dimension reduction. • PCA combined RFR can detect 1 % CO-RBD in VCO.