Binary classification of multifloral honeys using hybrid ensemble optimization via genetic algorithm based on physicochemical parameters

Reliable screening and discrimination between multifloral and monofloral honeys are important for supporting honey quality assessment, consumer confidence, and market transparency. In this study, the potential of physicochemical parameters combined with machine learning techniques was investigated for the binary classification of multifloral and monofloral honeys. A comprehensive dataset including multiple physicochemical characteristics such as electrical conductivity, moisture content, carbohydrate composition, colour intensity, and isotopic parameters was analyzed. To identify the most informative variables, the Minimum Redundancy Maximum Relevance (MRMR) feature selection method was applied. The results indicated that electrical conductivity, major sugar-related variables (F + G and glucose), stable carbon isotope parameters (δ¹³C protein and Δδ¹³C), and selected fold-specific minor sugars were among the most informative parameters for binary classification. Several machine learning algorithms, including Random Forest, neural network models, and deep learning approaches, were evaluated to assess their classification performance. In addition, a hybrid ensemble optimization framework based on a Genetic Algorithm (HEO-GA) was proposed to combine heterogeneous classifiers and improve predictive accuracy. The performance of the proposed method in the binary classification of multifloral and monofloral honeys was evaluated using standard metrics, including accuracy, F1-score, sensitivity, and specificity. The experimental results demonstrated that the proposed hybrid model achieved a highly balanced classification performance, comparable to that of Random Forest, while outperforming other baseline models. These results indicate that integrating physicochemical analysis with machine learning provides a potentially useful and scalable framework for screening and binary classification of multifloral and monofloral honeys. Furthermore, the contributions of the selected features to the binary classification of multifloral and monofloral honeys were interpreted using SHAP values and their corresponding distribution plots. The proposed framework was specifically developed to distinguish multifloral honeys from monofloral honeys. The findings highlight the potential of data-driven analytical approaches as complementary tools for quality control and the binary classification of multifloral and monofloral honeys.