Integrated chemometric and machine learning approaches to study the properties of pure and adulterated edible oils.

The adulteration of pure edible oils, particularly with cost-effective oils like palm oil, has become a significant concern due to its detrimental impact on oil quality and human health. This study examines how palm oil adulteration affects the dielectric, physical, and chemical properties of groundnut and sunflower oils. Additionally, this study explores the nutritional differences between pure and adulterated oils, highlighting potential risks to consumers. Microwave analysis showed decreased dielectric constant and loss in groundnut and sunflower oils after palm oil blending. Chemical parameters and fatty acid composition confirmed adulteration effects and potential health risks. To ensure the accuracy and reliability of our findings, we applied several chemometric methods, including Hierarchical Cluster Analysis (HCA), Principal Component Analysis (PCA), Multiple Linear Regression (MLR), and Artificial Neural Networks (ANN) to detect adulteration in oils. Among these, ANN and MLR were compared for predicting the dielectric constant. The results showed that ANN performed much better than MLR, explaining R² value of 0.94 in groundnut oil and 0.96 in sunflower oil, proving it to be a more accurate method for assessing oil quality. The features that made the ANN model work better were found using SHAP analysis. The result of the SHAP value shows that the refractive index (RF) in groundnut oil and the saponification (SAP) value in sunflower oil are the most influential predictors of dielectric constant, as both parameters vary significantly with adulteration. This finding demonstrates a powerful and accurate approach for detecting adulteration and assessing the quality of edible oils.