Comprehensive examination and comparison of machine learning techniques for the quantitative determination of adulterants in honey using Fourier infrared spectroscopy with attenuated total reflectance accessory

Abstract

Facile, robust, and accurate analyses of honey adulterants are required in the honey industry to assess its purity for commercialization purposes. A stacked regression ensemble approach using Fourier transform infrared spectroscopic method was developed for the quantitative determination of corn, cane, beet, and rice syrup adulterants in honey. A training set (n=81) was used to predict the percent adulterant composition of the aforementioned constituents in an independent test set (n=32). A comprehensive comparison of the performance of various machine learning techniques including support vector regression using linear function, least absolute shrinkage and selection operator, ride regression, elastic net, partial least squares, random forests, recursive partitioning and regression trees, gradient boosting, and gaussian process regression was assessed. The predictive performance of the aforementioned machine learning approaches was then compared with stacked regression, an ensemble learning technique which collates the performance of the various abovementioned techniques. Results show that stacked regression did not primarily outperform other techniques across all four syrup adulterant constituents in the testing set data. Further, elastic net generalized linear model generated the optimum results (Rootmeansquareerrorofprediction(RMSEP)average=0.0107,Raverage2=0.809) across all four honey adulterant constituents. Elastic net coupled with Fourier transform infrared spectroscopy may offer a novel, direct, and accurate method of simultaneously quantifying corn, cane, beet, and rice syrup adulterants in honey.