Comprehensive electrochemical and machine learning-based study of rancidity in four edible oils over various storage periods

Abstract

This study investigates the rancidity development in four edible oils (corn, mustard, soybean, and sunflower) over a 12-month storage period using a novel approach combining electrochemical techniques and machine learning. Cyclic voltammetry, electrochemical impedance spectroscopy, and differential pulse voltammetry were employed to characterize oil oxidation. Electrochemical parameters showed strong correlations with traditional chemical indicators, such as the DPV peak current at +0.2 V with p-anisidine value (r = 0.94, p < 0.001). A Random Forest model, trained on electrochemical data, accurately predicted Total Oxidation (TOTOX) values, achieving an R² of 0.96 and RMSE of 2.18 for the test set. The model effectively captured oxidation trends across oil types, with the highest accuracy for mustard oil (MAE: 1.21) and lower performance for sunflower oil (MAE: 2.15). Feature importance analysis revealed charge transfer resistance and DPV peak currents as the most influential predictors. This approach offers rapid, non-destructive assessment of oil quality, potentially improving quality control in the food industry. However, challenges such as electrode fouling and complex sample preparation need to be addressed for practical implementation.