Low-cost rapid workflow for honey adulteration detection by UV–Vis spectroscopy in combination with factorial design, response surface methodology and supervised machine learning classifiers

Abstract

Honey is often adulterated and heat treated for higher profits. This study aims to develop a low-cost, easy to perform, and less time-consuming method for screening the honey status. Known (12: adulterated 4; unadulterated 8) and unknown (7) honey samples of different geographical and biological sources were used to simulate adulteration and heat treatment effects by factorial design. Crude methanol-chloroform extracts of simulated samples were used for generating UV–Vis spectra (200-700 nm) and the effects were assessed by response surface methodology. Four machine learning classifiers were applied to predict the sample status, and among them Neural network and Random-forest were found to be satisfactory. The selected classifiers successfully detected 2 adulterated, and 4 unadulterated honeys among 7 unknown samples. Present workflow is proved to be an effective technique in detecting honey adulteration, and can be applied in honey and other food industries for optimization and quality control of their products.