Detection of adulteration in pure olive oil using the UV-Vis spectroscopy coupled with machine learning

This study employed ultraviolet-visible (UV-Vis) spectroscopy as a rapid, non-invasive analytical technique to detect adulteration in olive oils. Various proportions (ranging from 5% to 50%) of sunflower, soybean, and palm olein oils were blended with extra virgin, virgin, and refined olive oils to prepare adulterated samples. Transmission spectra within the 190–1100 nm range were collected to develop both predictive and classification models. Classification models, including SVM and DA, were utilized to differentiate pure and adulterated oil samples, while regression models, specifically PLSR and SVR, were used to predict adulteration rates. To improve data quality, preprocessing techniques such as SG, SNV, D1st, and DT were applied. The performance of the regression models was evaluated using R² and RMSE, whereas classification model performance was assessed based on accuracy. The results indicated that applying D1st and DT preprocessing significantly improved the calibration and validation performance of both PLSR and SVR models. The PLSR model, following D1st preprocessing, achieved the highest predictive accuracy, with R² values exceeding 0.9970 and RMSE values below 2%. Among the classification models, RBF-SVM achieved classification accuracies of 97.37% for sunflower oil, 98.7% for soybean oil, and 94.74% for palm olein oil. Similarly, the QDA model yielded classification accuracies of 98.68%, 100%, and 98.68% for sunflower, soybean, and palm olein oils, respectively. Overall, the findings highlight the high potential of integrating UV-Vis spectroscopy with machine learning algorithms as a precise and efficient approach for the rapid detection of adulteration in edible oils.