Rapid Identification of Orange Juice Adulteration Using Voltammetric Profiling and Machine Learning

บทคัดย่อ

In this study, the differential pulse voltammetry with a gold electrode and machine learning was employed to detect adulteration in orange juices. The method assessed both natural and commercial juices, along with their mixtures containing known proportions of natural juice. Initially, an unsupervised machine learning algorithm, Cluster Analysis (CA), was used to highlight differences, demonstrating the ability to distinguish between natural and flavored orange juices. Subsequently, supervised machine learning methods, including Interval Partial Least Squares – Linear Discriminant Analysis (iPLS-LDA) and Interval Partial Least Squares – Random Forest (iPLS-RF), were applied for classification purposes. The RF model achieved up to 95% classification accuracy, greatly exceeding 67.5% of iPLS- LDA. This enables reliable detection of orange juice adulteration. The RF model struggled to accurately distinguish between the “Natural” and “Mixed” categories, particularly for samples containing a medium proportion of natural orange juice (around 45–50%). The integration of voltammetric fingerprints with machine learning enabled a fast, cost-effective classification method for on-site analysis with portable sensors. This approach proved more efficient than other complex analytical techniques.