Computational intelligence applied to discriminate bee pollen quality and botanical origin

Area of Science:

• Agricultural Science
• Computational Intelligence
• Cheminformatics

Background:

• Bee pollen composition varies significantly based on botanical origin.
• Predicting pollen's botanical origin from its physicochemical properties is challenging.
• Computational intelligence offers potential solutions for complex biological data analysis.

Purpose of the Study:

• To develop and compare computational intelligence models for predicting bee pollen's botanical origin.
• To assess the models' ability to predict physicochemical composition from botanical origin.
• To identify the most effective model for each prediction task.

Main Methods:

• Development of predictive models using neural networks (NN), fuzzy models (FM), and support vector machines (SVM).
• Training models to predict predominant, secondary, and tertiary plant genera from physicochemical data.
• Training inverse models to predict physicochemical characteristics from known botanical origins.

Main Results:

• A probabilistic NN model achieved 98.4% accuracy in classifying the predominant plant genus of bee pollen.
• Models showed lower accuracy for predicting secondary and tertiary plant genera.
• Fuzzy models demonstrated excellent performance in predicting physicochemical characteristics, with prediction errors under 10%.

Conclusions:

• Computational intelligence models, particularly NN and FM, are effective for analyzing bee pollen composition.
• Accurate prediction of predominant botanical origin is feasible using physicochemical data.
• Fuzzy models offer high accuracy for predicting physicochemical properties from botanical origin.