Laser Raman Spectroscopic Assessment of Honey Adulteration by Molasses

Abstract

Honey adulteration by cheaper sweeteners such as sugar syrups, synthetic honey, molasses, and sugar beet has become a common vice thus negatively affecting the quality of honey production, and diminishing its market value. Lack of label–free, easy to use and rapid quality assessment honey adulteration detection techniques in the market has encouraged honey producers and processors to cheat on its quality. Furthermore, the current honey adulteration detection techniques such as, Stable Carbon Isotope Ratio Analysis (SCIRA), Liquid Chromatography (LC), Gas Chromatography (GC), and High Performance Liquid Chromatography (HPLC) suffer from the disadvantages that include being less rapid and expensive to use. Hence the need for rapid and affordable honey adulteration detection techniques. In this research, laser Raman spectroscopy robustness as an emergent technique for definitive molecular fingerprint analysis was explored to study honey adulteration. Authentic honey was intentionally adulterated by molasses in varying concentration ranges. Raman spectra was collected separately with each done under 60 seconds from small quantities of 1 g of authentic honey, molasses and molasses - adulterated honey samples. PCA was employed to perform exploratory analysis of the combined authentic and adulterated Raman spectral data sets, while machine learning techniques namely, random forest (RF), and support vector machine (SVM), and artificial neural networks (ANN) were used to create multivariate classification and regression models for forecasting authentic honey and molasses - adulterated honey samples. The most variant bands between authentic honey and molasses that were confirmed using ANOVA and PCA showed characteristic bands centered around: 690 cm-1 (stretching of CO and CCO, and bending of OCO); 732 cm-1 (glucose ν(C-C) vibrations); 754cm-1 (weak (C = O) bond vibrations); 845 cm-1 (glucose spectrum); 970 cm-1 (glucose, ν (C- O) vibrations). Furthermore, high classification accuracies ranging from 86 – 100 % were achieved using RF and SVM classification models. Artificial Neural Networks (ANN) was built as a regression model using the concentration ranges of 0 – 10%. The coefficient of determination (R –squared) was= 0.5786 and the mean absolute error (MAE) was 1.51. In order to calculate the limit of detection (LOD), the training data set obtained from the ANN regression model were used to determine the LOD. The median and mean absolute deviation values of the samples with known concertation versus samples whose concertation were predicted were used to calculate the LOD because they were found to be statistically stable, thus they yielded minimized error bars. Using the ANN model an LOD value lower than 1% was obtained. Thus, the results discussed in this research demonstrate the capability of Raman spectroscopy coupled with PCA, RF, and SVM, and ANN for molecular distinction of authentic and molasses - adulterated honey using the Raman spectral data.