Machine learning-assisted Raman spectroscopy for non-destructive analysis of crude palm oil quality

Abstract Quality assessment of crude palm oil remains a critical challenge globally, particularly in resource-poor areas where traditional methods are time-consuming and destructive. This study explores machine learning-assisted Raman spectroscopy for non-destructive assessment of peroxide value (PV) and iodine value (IV) in palm oil. Raman spectra were collected from 200 samples from five Ghanaian markets, with second derivative preprocessing significantly enhancing feature resolution. Twelve predictive models were developed by combining three variable selection algorithms (CARS, GA, UVE) with three regression methods (PLS, SVM, RF). The genetic algorithm-random forest (GA-RF) model demonstrated exceptional prediction accuracy for both PV (Rp = 0.9831, RPD = 7.7397) and IV (Rp = 0.9752, RPD = 6.3927). Key spectral regions associated with unsaturation (1287-1657 cm⁻¹) and oxidation (1748-1840 cm⁻¹) were identified as crucial predictors. This approach enables rapid, non-destructive quality assessment with potential applications throughout the palm oil value chain.