Machine learning strategy for light lamb carcass classification using meat biomarkers

Abstract

In Mediterranean areas, lamb meat is considered to be of great commercial value. Moreover, consumers are becoming increasingly interested in understanding the origin of lamb meat and its associated production and breeding systems. Among many applications, algorithms based on artificial intelligence are used to identify the origin of food products, and in this context, algorithms such as the Support Vector Machine (SVM), K-Nearest Neighbours (KNN), and the Artificial Neural Network (ANN) have been proposed to differentiate the origin of the animals according to their feeding diet. The objective of this study was to evaluate the performance of a variable reduction method based on a multiple regression model and three widely-used machine learning algorithms (SVM, KNN and ANN) for the classification of three commercial light lamb carcasses, from three feeding diets, in an indigenous Spanish breed (Mallorquina), using fatty acid and volatile compound biomarkers of meat. Machine learning algorithms were employed to discriminate lamb carcasses using 14 identified significant biomarkers, which were arranged based on an estimation of the relative importance (stepwise forward multiple regression F-score) of the input variables. We achieved high performances for the SVM, KNN and ANN algorithms, with 86%, 98% and 98% prediction accuracy, respectively. Among the 14 biomarkers used, 7 were identified as showing the highest discriminant capacity. The F-scores indicate that C17:1 and C20:5 n-3 fatty acids, and 2,5-dimethylpyrazine and 3-methylbutanal volatile compounds are the four most relevant biomarkers for predicting three lamb feeding diets.