Fecha de publicación: -- Fuente: PubMed "olive oil" Talanta. 2026 Sep 25;313(Pt C):130673. doi: 10.1016/j.talanta.2026.130673. Online ahead of print.ABSTRACTDetection of adulteration in extra-virgin olive oil (EVOO) and tracing possible sources remain critical challenges for food authenticity, consumer safety, and regulatory enforcement, particularly in resource-limited settings where the application of high-accuracy analytical tools is impractical. In this work, a practical integration of a machine learning system with a compact, ultra-low-cost smartphone-based prompt fluorescence imaging platform is presented for on-site detection of EVOO adulteration, where prompt fluorescence refers to fluorescence recorded during excitation without time-delayed detection. The proposed platform employs a λ= 405 nm laser excitation source and orthogonal fluorescence imaging geometry, customized with a smartphone CMOS camera. The smartphone runs as a two-stage analytical framework, comprising adulterant classification followed by concentration estimation. The performance of machine-learning approaches (LightGBM, SVM, KNN, RF, Logistic Regression, Naive Bayes, Gradient Boosting, Decision Tree, Lasso Regression, SVR) has been evaluated using EVOO adulterated with common sources of adulteration such as refined olive oil, soybean oil, and palm oil over a wide concentration range. Machine learning models attain strong discrimination ability and high predictive accuracy by utilizing non-linear relationships within the fluorescence-derived RGB features. Independent validation using real samples confirms reliable adulterant identification and accurate EVOO purity estimation, with an average error of 3.1%. The present framework has been developed and validated using binary mixtures containing EVOO and a single adulterant, with the minimum operational detection threshold being 2% adulterant concentration. The proposed approach overcomes key limitations of conventional methods by enabling simultaneous detection of adulterant type and quantification without prior knowledge of the adulterant type, and therefore allowing users to predict the possible source of adulteration.PMID:42804948 | DOI:10.1016/j.talanta.2026.130673