Machine Learning in Sensory Analysis of Mead-A Case Study: Ensembles of Classifiers

Area of Science:

• Food Science and Technology
• Computational Chemistry
• Data Science

Background:

• Classifying mead varieties based on sensory attributes and volatile organic compounds is complex.
• Machine learning offers advanced analytical capabilities for intricate datasets in food science.
• Exploratory data analysis is crucial for identifying key features in mead characterization.

Purpose of the Study:

• To explore the application of machine learning techniques for mead classification.
• To identify characteristic sensory features and aromatic compounds of different mead types.
• To evaluate the performance of various machine learning algorithms in mead classification.

Main Methods:

• Utilized cluster mapping and k-means clustering for exploratory analysis of mead features.
• Employed machine learning algorithms including Random Forest, AdaBoost, Bagging, KNN, and Decision Tree for classification.
• Performed error matrix analysis to assess algorithm performance and identify misclassifications.

Main Results:

• Random Forest and K-Nearest Neighbors algorithms demonstrated high accuracy in mead classification.
• The Decision Tree algorithm achieved the highest accuracy (0.909) for classification based on aroma.
• Acacia mead was identified more easily by algorithms compared to tilia or buckwheat mead.

Conclusions:

• Combining exploratory methods (cluster map, k-means) with machine learning enhances mead classification.
• Algorithm selection and optimization are critical for successful mead identification.
• Machine learning provides a powerful framework for objective mead characterization.