Detection of honey adulteration using machine learning

Area of Science:

• Food Science
• Analytical Chemistry
• Computer Science

Background:

• Honey adulteration is a significant global concern, impacting consumer health and market integrity.
• Traditional detection methods, such as Melissopalynology, lack the sensitivity and efficiency required for modern quality control.
• Hyperspectral imaging offers a non-destructive and rapid approach for food quality assessment.

Purpose of the Study:

• To comparatively evaluate machine learning algorithms for detecting honey adulteration.
• To develop an accurate and efficient honey counterfeit detection technology.
• To improve existing models through hyperparameter tuning for enhanced classification accuracy.

Main Methods:

• Utilized hyperspectral imaging to capture spectral data from honey samples.
• Employed various machine learning algorithms including Artificial Neural Networks (ANN), Support Vector Machines (SVM), K Nearest Neighbors (KNN), Random Forests, and Decision Trees.
• Applied feature reduction techniques like feature ranking and autoencoders for data processing and classification.
• Pre-processed and segmented hyperspectral images of adulterated honey samples from multiple brands and botanical origins.

Main Results:

• Achieved over 98% classification accuracy in identifying adulterated honey samples.
• Identified C1 Clover honey as the most frequently misclassified sample type.
• Demonstrated the effectiveness of machine learning combined with hyperspectral imaging for honey authentication.

Conclusions:

• Machine learning and hyperspectral imaging provide a robust solution for detecting honey adulteration.
• The developed technology offers high accuracy and efficiency, surpassing traditional methods.
• Further research can refine models to address challenges like generalization to unknown honey types and reduce misclassification rates.