Comparison of Genetic and Non-genetic Algorithm Partial Least Squares for Sugar Quantification in Philippine Honeys

Quantification of sugars is considered to be a key factor in the assessment of honey quality. Standard chromatographic methods for sugar quantification in honey may be unsuitable for routine analysis due to the associated time, cost, and complexity. Herein, a comparison between non-genetic and genetic algorithm-partial least squares (GA-PLS) chemometric and infrared spectroscopic approaches is reported for the direct quantification of glucose, fructose, and sucrose in Philippine honeys. A full factorial set (n = 61) was designed to predict Box-Behnken test set (n = 16) sugar concentrations. First derivative (FD) signal processing followed by PLS of the 1500-800 cm(-1) spectral region harboring the analytes provided optimum results with R-average(2) = 0.943 in the training set and a mean root mean square error of prediction (RMSEPaverage) of 0.049 across fructose, glucose, and sucrose. Applying GA-PLS after FD processing across the spectral region from 3700 to 700 cm(-1) did not further improve the results (R-average(2) = 0.926, RMSEPaverage = 0.077). FD followed by PLS (1500-800 cm(-1)) provided minimal differences between the actual and predicted sugar content in training and testing values, implying that the probe produced developmental avenues to directly quantify these sugars. When applied to real samples (n = 26), the method potentially differentiated real honey from non-honey and adulterated samples.