Food Authenticity Models for Mytilus galloprovincialis (Mediterranean Mussel): Exploratory Study.

Area of Science:

• Marine Biology
• Food Science
• Analytical Chemistry

Background:

• Geographical origin determination is crucial for seafood fraud prevention, food safety, and sustainable resource management.
• Mussel origin verification is complex due to global farming practices and trade.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) models combined with trace element fingerprinting (TEF) and stable isotope ratio analysis (SIRA) for determining mussel geographical origin.
• To assess the predictive accuracy of different ML algorithms (random forests, support vector machines, artificial neural networks) using elemental and isotopic data.

Main Methods:

• Mussel shells and tissues were analyzed for 14 trace elements and carbon/nitrogen isotope ratios.
• Data from eight global regions and ten locations were used to train and test ML models.
• Models were developed using individual or combined datasets, including variable selection via random forest.

Main Results:

• Random forest and artificial neural network models achieved 100% accuracy in predicting mussel region and location.
• Trace element fingerprinting (TEF) data, particularly when combined with ML, proved highly effective.
• Stable isotope ratio analysis (SIRA) models showed lower prediction accuracies compared to TEF.

Conclusions:

• Machine learning models, especially random forests and artificial neural networks, are highly effective for determining mussel geographical origin.
• TEF combined with ML offers a robust solution for mussel traceability, enhancing food safety and sustainability.
• This approach has significant implications for combating seafood fraud and supporting global aquaculture management.