Classification of meat species in sausage products using FTIR spectroscopy and multivariate analysis for food authentication

The growing concern about food fraud and mislabeling has created a need for reliable and efficient methods to identify meat species in processed products. This study introduces a non-destructive approach using Fourier Transform Infrared (FTIR) spectroscopy to classify beef, chicken, donkey, and horse meat in sausage formulations. A total of 80 sausage samples were analyzed (20 samples per species). FTIR spectral data underwent preprocessing and dimensionality reduction prior to classification with four machine learning models. Among these, the Random Forest model achieved the best overall performance, with 100 % accuracy and a macro F1-score of 1.00 (95 % CI [1.00–1.00]) on the independent test set. Logistic Regression and K-Nearest Neighbors also achieved perfect test performance, while Support Vector Machine reached 96 % accuracy (95 % CI [0.88–1.00], F1 = 0.95). These findings demonstrate the feasibility of FTIR for rapid meat species identification and offer a practical tool for regulators and industry stakeholders to enhance consumer trust and combat food fraud.