DETECTION AND QUANTIFICATION OF COCOA POWDER ADULTERATION USING NIR SPECTROSCOPY

Resumen

Cocoa powder has many uses and benefits, increasing demand on the market followed by an increase in the price of cocoa powder. This occurrence leads to the adulteration of cocoa powder and thus degrades the quality. NIR spectroscopy with multivariate analysis was applied to assess cocoa powder adulteration. This study aimed to detect and quantify cocoa powder adulteration using classification and regression models. The adoption of partial least square, Ridge, Lasso, Elastic Net, and random forest (RF) regression models was also evaluated to quantify adulterants carob powder, cocoa shell powder, foxtail millet flour, soybean powder, and whole wheat flour in cocoa powder. The NIR spectroscopy spectra show the overlapped within all samples, while after the Savitzky-Golay (SG) smoothing, some wavelengths have different absorption at therange 1650 – 1700 nm, 2030 – 2080 nm, 2225 – 2290 nm, and 2430 – 2500 nm. Unsupervised techniques (principal component analysis, PCA; hierarchical component analysis, HCA) show the clear classification trend of the samples based on the type and percentage of adulteration. The application of the supervised technique RF and support vector machine can classify samples with 100% accuracy. RF also selects the importantwavelength to classify samples based on the type of adulterants. Quantifying the percentage of the sample adulteration using 5 model regression provides the result of R2 above 0.98 and root mean square error (RMSE) less than 1.7 for the test set, except for the RF regression model. The application of Boruta algorithm delivers better results for some models. The best regression model for quantification was PLS-Regression with Boruta algorithm with R2 of 1 and RMSE of 0.0002 for the test set. Accordingly, NIR spectroscopy with multivariate analysis can detect and quantify adulterated cocoa powder.