Geographical classification of wine and olive oil by means of classification and influence matrix analysis (CAIMAN)

Abstract

Classification and influence matrix analysis (CAIMAN) is a new classification method, recently proposed and based on the influence matrix (also called leverage matrix). Depending on the purposes of the classification analysis, CAIMAN can be used in three outlines: (1) D-CAIMAN is a discriminant classification method, (2) M-CAIMAN is a class modelling method allowing a sample to be classified, not classified at all, or assigned to more than one class (confused) and (3) A-CAIMAN deals with the asymmetric case, where only a reference class needs to be modelled. In this work, the geographic classification of samples of wine and olive oil has been carried out by means of CAIMAN and its results compared with discriminant analysis, by focusing great attention on the model predictive capabilities. The geographic characterization has been carried out on three different datasets: extra virgin olive oils produced in a small area, with a “protected denomination of origin” label, wines with different denominations of origin, but produced in enclosed geographical areas, and olive oils belonging to different production areas. Final results seem to indicate that the application of CAIMAN to the geographical origin identification offers several advantages: first, it shows – on an average basis – good performances; second, it is able to deal in a simple way classification problems related to tipicity, authenticity, and uniqueness characterization, which are of increasing interest in food quality issues.