Researchers evaluate AI effectiveness in predicting food safety alerts

A study published in Scientific Reports has examined the ability of machine learning models to predict emerging international food safety notification links. By analyzing 28,724 notifications from the European Union’s Rapid Alert System for Food and Feed (RASFF) between 2019 and 2025, researchers assessed whether network-based models could help authorities anticipate unreported contamination pathways.

The investigation found that while directional modeling improved predictions compared to conventional undirected models, the addition of multiplex layering and graph neural networks provided negligible predictive value. Furthermore, the study noted that standard evaluation schemes often produced misleadingly high performance scores, highlighting a need for more rigorous comparisons with randomized networks when designing early-warning surveillance systems.

In a separate context regarding European food safety, the European Food Safety Authority (EFSA) conducted a pesticide-residue assessment on 9,842 samples collected in 2024. The assessment found that 98.8% of the samples complied with EU legislation, while 2.4% had residues exceeding maximum levels and 1.2% were confirmed as non-compliant after accounting for measurement uncertainty.