Sensing fermentation degree of cocoa (Theobroma cacao L.) beans by machine learning classification models based electronic nose system

Abstract

Abstract Cocoa bean fermentation is an important postharvest process that develops aroma and processing properties. Although the cocoa fermentation is of high complexly, farmers are employing empirical methods to determine the fermentation degree of cocoa. Researchers, on the other hand, are using expensive equipment such as high‐performance liquid chromatography and gas chromatography‐mass spectrometry to study cocoa fermentation. In this study, machine learning based electronic nose system, a fast measuring and affordable method, was developed to determine the fermentation degree of cocoa beans. Six machine‐learning methods (bootstrap forest, boosted tree, decision tree, artificial neural network (ANN), naïve Bayes, and k‐nearest neighbors) were conducted to classify the fermentation time of cocoa beans. Bootstrap forest algorithm achieved a misclassification rate as low as 9.4%. ANN and boosted tree achieved 12.8 and 13.6% misclassification rate respectively. However, other methods failed to do classification for cocoa beans. Practical applications The electronic nose system can be used by cocoa farmers to monitor cocoa fermentation and ensure the quality of cocoa beans. The method is relatively inexpensive and easy to operate.