Step by step approach to multi-element data analysis in testing the provenance of wines

Abstract

The possibility to classify the provenance of a wide variety of randomly selected wines according to multi-element analysis data was tested. Large number of parameters is used for solution of such complex problem and the role of the noise increases. Stepwise approach is tested dividing the wine origin classification into some steps to simplify the problem. Outcomes of the approach are studied on the basis of the chemical analytical data obtained for 23 elements in 103 wines from seven countries. Anova was used to select the most informative elements at each step. Three or four elements often were found sufficient to discriminate between countries at 0.9 probability level. Principal component analysis was applied for concise data presentation. The possibility of application of the multivariate normal distribution to the principal components was tested and confirmed, and thoroughly used for the classification power estimates. Problems of indication of batches and adequate representation of those by samples are emphasized.