Classification and identification of chicken-derived adulteration in pork patties: A multi-dimensional quality profile and machine learning-based approach

Abstract

The purpose of this study was to identify pork patty samples with different levels of chicken adulteration utilizing machine learning techniques. A multi-dimensional detection system was constructed using 300 samples comprising 0 %–100 % chicken adulteration gradients. Forty-three parameters, including water activity, color, texture properties, free amino acids, and electronic nose sensor responses, were analyzed. The physical and chemical changes in pork patties caused by adulteration were systematically analyzed. Machine learning models, such as partial least squares discriminant analysis (PLS-DA), support vector machines (SVM), and back-propagation artificial neural networks (BP-ANN), were developed and evaluated based on quality indices and adulteration rates. The BP-ANN model achieved 99.52 % training accuracy, with an independent test set accuracy of 98.00 %. This performance was significantly superior to that of SVM (95.56 %) and PLS-DA (86.75 %). Shapley Additive Explanations (SHAP) analysis identified threonine content (Thr), chroma parameter (C∗), and histidine content (His) as the primary discriminant indices. The six screening features (L∗, C∗, S12, Thr, Lys, His) combined with the electronic nose sensor array S5, S11, S13, S14 collectively enhanced model robustness. The BP-ANN algorithm was integrated into the Streamlit platform to develop a real-time online tool for predicting adulteration rates.