Multi-source data fusion for soybean origin traceability: Stable isotopes, elemental composition, & volatile organic...

Area of Science:

• Agricultural Science
• Food Science
• Data Science

Background:

• Global soybean demand has doubled, increasing risks of food fraud, deforestation, and climate change.
• The European Union Deforestation Regulation (EUDR) mandates high-resolution traceability for sustainable soybean sourcing.
• Integrating multiple analytical methods via data fusion is essential for accurate origin tracing.

Purpose of the Study:

• To evaluate the effectiveness of four data fusion strategies for soybean traceability.
• To assess the performance of Low-level, Mid-Principal Component Analysis-Random Forest (PCA-RF), Mid-Uniform Manifold Approximation and Projection-Random Forest (UMAP-RF), and High-level fusion.
• To determine the optimal data fusion approach for meeting regulatory and sustainability demands.

Main Methods:

• Analysis of 60 soybean samples from six Brazilian states.
• Utilized stable isotope analysis, elemental profiling, and volatile organic compound characterization.
• Implemented and compared four distinct data fusion strategies: Low-level, Mid-PCA-RF, Mid-UMAP-RF, and High-level.

Main Results:

• High-level data fusion achieved 100% classification accuracy on the test set.
• Mid-UMAP-RF demonstrated strong performance with 99% classification accuracy.
• Data fusion significantly improved the accuracy of tracing soybean origins.

Conclusions:

• Data fusion is a powerful tool for enhancing soybean traceability.
• The study validates the effectiveness of advanced fusion techniques in meeting regulatory requirements like the EUDR.
• Improved traceability supports sustainable agricultural practices and mitigates environmental and economic risks.