SEARCH

SEARCH BY CITATION

Keywords:

  • Dimension reduction;
  • extra virgin olive oil;
  • interpretation of neural network results;
  • protected denominations of origin

Summary

A computerised approach to vastly reduce the experimental information required (number of independent variables) to classify similar extra virgin olive oils (EVOOs) is presented. It is based on the application of a multilayer perceptron (MLP) and further analysis of the obtained results using differential calculations. To validate this new model, it has been applied for the classification of 147 EVOO samples into four similar families. The oil samples employed came from two types of protected denomination of origin (PDO) oils and two non-PDO from the same Spanish province (Granada). This approach results in a new method that reduces the necessary size of the databases used, without an appreciable loss of information, by over 82%. The percentage of misclassifications using less data points is similar to the results achieved using the whole database (less than 0.90%).