Phenolic Profile as a Powerful Machine Learning Tool for Identification, Traceability, and Quality Control of Olive Cultivars

This study investigates the phenolic and fatty acid profiles of olives from four Olea europaea cultivars (Arbequina, Arbosana, Frantene, and Koroneiki), widely grown in the Mediterranean region and collected at different ripening stages in Italy. The aim was to assess the potential of olive chemical profiles as markers for cultivar classification using machine learning algorithms, including Linear Discriminant Analysis (LDA), K-Nearest Neighbors (KNN), Naive Bayes (NB), Random Forest (RF), and Support Vector Machine (SVM). Results showed that phenolic profiling achieved significantly higher classification accuracy than fatty acids across all models. Using phenolic data, RF and SVM achieved 98% accuracy, while for fatty acids, the best-performing model was NB, reaching just 65%. Importantly, minor phenolic compounds such as chlorogenic acid, ferulic acid, and apigenin were crucial for classification. These findings demonstrate that machine learning combined with phenolic profiling can improve cultivar identification, traceability, and quality control in the olive sector.