Multivariate and machine learning approach to honey adulteration: Integrating Principal Component Analysis and artificial neural network models

Honey adulteration is a major issue in India. It affects product quality and can pose health risks to consumers. This study examined the concentration of adulterants' impacts on honey quality using four types from different floral sources. The samples were mixed with brown rice syrup, corn syrup, and malt syrup at levels of 10%, 20%, and 30%. We analyzed nine physicochemical properties: moisture content, color, pH, electrical conductivity, density, viscosity, total soluble solids (TSS), hydroxymethylfurfural (HMF) content, and diastase activity. The results showed moisture content ranging from 15.23% to 18.00%, pH between 3.77 and 4.20, color ΔE from 20.71 to 29.71, TSS between 75.07°Bx and 80.83°Bx, electrical conductivity from 0.20 to 0.32 mS/cm, viscosity ranging from 5.51 to 16.80 Pa·s, density from 1.35 to 1.44 kg/m³, HMF content between 17.24 and 21.57 mg/kg, and diastase activity from 10.74 to 13.56. The analysis of variance (ANOVA) showed significant increases in HMF content and pH with the addition of adulterants. At the same time, viscosity, diastase activity, and electrical conductivity decreased significantly (P<0.05). Principal Component Analysis (PCA) found three main components that accounted for 83.5% of the variability in the data. Electrical conductivity, pH, color, and diastase activity were the key factors that distinguished the samples. An Artificial Neural Network (ANN) model using feed-forward backpropagation showed high accuracy (R²=0.96 for training, R²=0.94 for testing). It highlighted HMF content, pH, and diastase activity as the most important parameters for detecting adulteration. This combined multivariate machine learning approach offers a strong method for assessing honey quality and detecting adulteration