Abstract:
Fermentation level is a key bean quality indicator in the cocoa industry. Colorimetric sensor e-nose (CS e-nose) and an
innovatively designed near infrared chemo-intermediary-dyes spectra technique (NIR-CDS) combined with four
chemometric algorithms – extreme machine learning (ELM), support vector machine (SVM), linear discriminant analysis
(LDA) and k- nearest neighbor (k-NN), were applied to classify 90 sampled cocoa beans into three quality grades – fully
fermented, partially fermented and non-fermented. CS e-nose (89% ≤ Rp ≤ 94%) and NIR-CDS (85% ≤ Rp ≤ 94%) achieved
comparable classification rates; with the systems data cluster analysis yielding cophenetic correlation coefficients of (0.85 -
0.89). Both systems combined with SVM and ELM achieved high classification rate (Rp = 94%) and could be applied to
cocoa bean quality classification on in-situ and nondestructive basis. This novel NIR-CDS technique proved a pragmatic
approach for the selection of sensitive chemo-dyes used in the fabrication of e-nose colorimetric sensor array compared
with the hitherto trial-and-error method, which is time-consuming and dye-wasteful. The technique could also be
deployed in near-infrared systems for the detection of volatile (gaseous) compounds, which previously had been a
limitation.
