Prediction of chocolate quality from near-infrared spectroscopic measurements of the raw cocoa beans

Abstract

A large number of samples of chocolate from a wide range of cocoa beans were evaluated by a trained sensory panel. Analysis of the sensory data showed clear differences between cocoa from different geographical areas and due to different processing of the beans. The near-infrared (NIR) spectra of raw and roasted beans, chocolate mass and finished chocolate which corresponded with the chocolate samples in the sensory study were obtained. Analysis of the NIR data indicated similar discrimination of the samples, this being most apparent with raw bean samples and least with finished chocolate. Sophisticated mathematical and statistical procedures were used to modify and relate the two data sets and a correlation of 0.86 was shown to exist between the sensory data on the finished chocolate and the NIR data for raw beans. A predictive model was developed from the NIR data on raw beans which aimed at differentiating low quality samples from acceptable quality samples. The model identified 64% of low quality samples, while rejecting 20% of acceptable quality samples. In view of the uncertainty inherent in all sensory data this result is considered to hold great promise for the replacement of difficult and demanding sensory analysis by simple and reliable measurement of NIR absorption for the evaluation of cocoa and other commodities which are traded for their sensory properties.