Class modeling techniques in the control of the geographical origin of wines

摘要: Abstract Wine samples of four different countries: Hungary, Czech Republic, Romania and South Africa, have been studied within the European project WINES-DB “establishing a wine data bank for analytical parameters from third countries”. For each country two types were collected, during three consecutive years: commercial wines obtained by microvinification according to EC regulation N. 2729/2000. The sampling design was organized represent both grape varieties official regions in countries. 1188 analyzed 58 chemical quantities. Data analysis performed with special attention real problem, namely control frauds. Class modeling techniques (UNEQ, SIMCA, MRM) applied, answer general question: “Does sample O, stated class A, really belong A?”. Two validation strategies, based on cross an external, representative, evaluation set, used evaluate carefully predictive performance models. results indicate that countries it is possible compute models high efficiency, generally reduced number variables. To obtain efficient models, red white wines, must be considered separately. validity ensured representativity samples, appropriate application Chemometrics validation.