Pattern recognition and genetic algorithms for discrimination of orange juices and reduction of significant components from headspace solid‐phase microextraction

Introduction – Orange (Citrus sinensis L.) juice comprises a complex mixture of volatile components that are difficult to identify and quantify. Classification and discrimination of the varieties on the basis of the volatile composition could help to guarantee the quality of a juice and to detect possible adulteration of the product. Objective – To provide information on the amounts of volatile constituents in fresh‐squeezed juices from four orange cultivars and to establish suitable discrimination rules to differentiate orange juices using new chemometric approaches. Methodology – Fresh juices of four orange cultivars were analysed by headspace solid‐phase microextraction (HS‐SPME) coupled with GC‐MS. Principal component analysis, linear discriminant analysis and heuristic methods, such as neural networks, allowed clustering of the data from HS‐SPME analysis while genetic algorithms addressed the problem of data reduction. To check the quality of the results the chemometric techniques were also evaluated on a sample. Results – Thirty volatile compounds were identified by HS‐SPME and GC‐MS analyses and their relative amounts calculated. Differences in composition of orange juice volatile components were observed. The chosen orange cultivars could be discriminated using neural networks, genetic relocation algorithms and linear discriminant analysis. Genetic algorithms applied to the data were also able to detect the most significant compounds. Conclusions – SPME is a useful technique to investigate orange juice volatile composition and a flexible chemometric approach is able to correctly separate the juices. Copyright © 2009 John Wiley & Sons, Ltd.