Functional Data Analysis and Machine Learning for the Traceability of Italian Sheep Milk

Abstract

Authenticating Protected Designation of Origin dairy products requires robust and rapid analytical tools. In this work, we analyze the potential of Fourier Transform InfraRed spectroscopy combined with Functional Data Analysis and Machine Learning to predict the altitude of production for sheep’s milk. We compare the predictive power of the entire spectral domain against specific sub-intervals selected via Sparse Functional Linear Discriminant Analysis. The results show that selecting discriminant sub-intervals improves accuracy. The models built on these sub-intervals reached at least 95% accuracy for both Valle del Belice and Sarda breeds, offering a non-destructive alternative to traditional chemical methods.