Applying machine learning to high-quality wine identification

This paper discusses a machine learning approach, aimed at the definition of methods for authenticity assessment of some of the highest valued Nebbiolo-based wines from Piedmont (Italy). This issue is one of the most relevant in the wine market, where commercial frauds related to such a kind of products are estimated to be worth millions of Euros. The main objective of the work is to demonstrate the effectiveness of classification algorithms in exploiting simple features about the chemical profile of a wine, obtained from inexpensive standard bio-chemical analyses. We report on experiments performed with datasets of real samples and with synthetic datasets which have been artificially generated from real data through the learning of a Bayesian network generative model.