Discrimination of thermally treated milk using fluorescence spectroscopy combined with PCA and artificial neural networks

This study assessed the use of multilayer perceptron artificial neural networks (MLP-ANN) combined with fluorescence spectroscopy to classify milk samples by species (cow or buffalo), thermal treatment, and cold storage duration. Models were developed using different fluorescence inputs: Vitamin A, tryptophan, synchronous spectra (80 nm offset), and a Vitamin A–tryptophan combination. Model performance was evaluated based on classification accuracy, mean square error (MSE), and training efficiency. Principal component analysis (PCA) was also employed to explore the effects of cooling, thermization, and pasteurization on milk’s physicochemical properties. PCA results showed clear clustering of samples based on species, thermal treatments, and storage times, with associated differences in pH, lactose, acidity, nitrogen fractions, calcium, and clotting time. MLP-ANN models achieved classification accuracies ranging from 44% to 100% and MSE values from 0% to 14%, depending on input features and classification complexity. Synchronous fluorescence spectra yielded the highest performance, achieving 100% accuracy in distinguishing both species and thermal treatments, highlighting their sensitivity to heat-induced molecular changes. Simpler classification tasks required fewer training epochs and hidden neurons, while more complex tasks involving multiple variables demonstrated lower performance, particularly when using combined tryptophan–Vitamin A inputs. • PCA clustered milk samples by species, treatment, and storage time based on physicochemical features. • Fluorescence spectroscopy has proved to be a reliable method to detect thermal treatments of milk. • MLP-ANN with fluorescence was useful to differentiate milk based on their storage time, thermal history and animal origin. • MLP-ANN with synchronous, Vit A or tryptophan spectra and their combinations gave 100% success classification rate for each milk classes.