Differential scanning calorimetry coupled with machine learning technique: An effective approach to determine the milk authenticity

Differential scanning calorimetry (DSC) coupled with machine-learning tools (random forest, gradient boosting machine, and multilayer perceptron, RF, GBM, MLP) were used to detect adulteration of raw bovine milk (formaldehyde, whey, urea, and starch). Adulterated samples presented a different DSC profile from raw milk; in particular the crystallization, melting and boiling points of cow milk were significantly affected by adding the adulterants starch, formaldehyde, cheese whey and urea while onset and peak temperatures of melting and boiling were especially affected by the incorporation of those adulterants. GBM and MLP were able to classify 100% of adulterated samples, whereas RF showed optimal performance with recognition and prediction capability of 100% and 88.5%, respectively. Overall, peak temperature of crystallization was the most important discriminating predictor for GBM and RF models, whereas peak temperature of boiling followed by onset temperature of crystallization and onset temperature of boiling were the most important predictors for MLP model. The detection of adulteration in milk has a multidimensional approach and DSC associated with machine-learning methods present an interesting perspective with practical potential to be adopted by the dairy industry. The potential application of fast measurements as DSC analysis coupled with machine-learning to determine the type of adulteration in milk can be a very promising tool for dairy industry to monitor the authenticity of milk and their use should be incentived for the dairy industry.