Abstract:
Groundnut oil (GNO) is a rich source of crucial fatty acids for human physiological development. However, concerns have been raised regarding certain manufacturers who may adulterate GNO with less expensive alternatives such as palm olein (PO). The authors proposed a robust and effective method that integrates Attenuated Total Reflection Mid-Infrared (ATR-MIR) spectroscopy with pre-processing, Principal Component Analysis (PCA), and Machine Learning (ML) models utilizing a stacking approach for the prediction and quantification of PO adulteration in GNO. Unlike earlier research, which mainly relied on single regression or classification models without thorough validation, this study uses both regression and classification approaches inside a stacking architecture to improve prediction resilience. This study used a dataset with pure groundnut oil with varying concentrations of palm oil (0 percent, 6.25 percent, 25 percent, and 50 percent). The results indicated that the stacking regressor (STR) achieved a coefficient of determination (R sup(2)) of 0.999, with a Root Mean Square Error (RMSE) of 0.145 ml (v/v), Standard Error of Prediction (SEP) of 0.006 ml (v/v), and Ratio of performance to deviation (RPD) of 136.2. The stacking classifier (STC) also attained a perfect accuracy rate of 100.0 percent. These outcomes show spectral-based ML approaches' effectiveness in food authentication, providing a non-destructive, quick, cost-effective, and adaptable solution for identifying adulteration in edible oils. Furthermore, this study employs mid-infrared spectral data to develop a hybrid framework for GNO adulteration detection.
