Automated detection of stale beef from electronic nose data

Area of Science:

• Food Science
• Analytical Chemistry
• Computer Science

Background:

• Accurate detection of stale beef is essential for consumer protection.
• Traditional methods for detecting spoilage can be time-consuming and subjective.
• Novel approaches are needed to objectively assess beef freshness.

Purpose of the Study:

• To develop and validate a method for accurately classifying stale and fresh beef samples.
• To compare the performance of various machine learning algorithms for beef spoilage detection.
• To establish a reliable system for identifying stale beef on the market.

Main Methods:

• Utilized an electronic nose to capture volatile organic compound profiles of beef samples.
• Employed linear discriminant analysis for dimensionality reduction of electronic nose data.
• Applied machine learning algorithms including extreme gradient boosting, logistic regression, K-nearest neighbor, random forest, support vector machine, and neural networks.
• Evaluated model performance using 10-fold cross-validation, F1 scores, and AUC values.

Main Results:

• All tested machine learning models achieved high accuracy (≥95%) in classifying beef samples.
• F1 scores and AUC values for most models exceeded 0.96.
• The support vector machine (SVM) algorithm demonstrated superior performance, achieving 100% accuracy with F1/AUC scores of 1.0.
• SVM effectively discriminated between fresh and stale beef based on electronic nose data.

Conclusions:

• Electronic nose technology combined with machine learning, particularly SVM, offers a highly accurate and reliable method for detecting stale beef.
• This approach provides a foundation for developing advanced, objective tools for food quality assessment.
• The findings pave the way for novel research in non-destructive detection of meat spoilage.