Classification of apple fruits according to their maturity state by the pattern recognition analysis of their polyphenolic compositions

Polyphenolic profiles of cider apple cultivars were studied in order to differentiate fruits according to their maturity state (ripe or unripe). Thiolysis and direct solvent extracts of freeze-dried apple pulps and peels were analysed by HPLC-DAD. Univariate data treatment did not achieve the mentioned target; thus a multivariate approach was considered. For each apple tissue data set, several chemometric techniques were applied to the most discriminant variables. Cluster analysis allowed a preliminary study of the data structure. Then, supervised pattern recognition procedures, namely linear discriminant analysis, K-nearest neighbours, soft independent modelling of class analogy, and multilayer feed-forward artificial neural networks (MLF-ANN), were used to develop decision rules to classify samples in the established categories. Excellent results were afforded by MLF-ANN applied to the concentrations of total procyanidins and (+)-catechin and the average degree of polymerisation of procyanidins in apple pulp, with success in the prediction ability of 97% and 99% for unripe and ripe categories, respectively.