Machine learning-driven intelligent transformation of food safety: from multi-source risk identification to...

The globalization of food supply chains has brought about increasingly complex food safety risks. Traditional approaches often have limitations such as low efficiency and poor adaptability when processing large-scale data, conducting real-time risk assessments, and analyzing multi-factor correlations, making it difficult to effectively manage these risks. As a data-driven intelligent technology, machine learning (ML) offers promising solutions by enabling efficient and accurate food safety management. This review systematically summarizes the progress of ML applications in several key domains, including food authenticity detection, toxicity prediction, assessment of chemical migration from packaging materials, and supply chain risk analysis and control. By integrating heterogeneous data from chemical, biological, and sensor-based sources, ML facilitates multi-source risk identification, providing support for comprehensive detection and prediction in these fields. Emphasis is placed on strategies such as optimizing data quality, improving model architectures, and establishing closed-loop experimental validations, which collectively enhance the accuracy and reliability of ML-based risk assessment frameworks. Nevertheless, challenges remain, particularly in terms of data availability and quality, model generalizability, computational requirements, and implementation costs. Future development will be driven by technological innovation, interdisciplinary integration, and the formulation of standardized evaluation frameworks, aiming to establish an intelligent and more resilient food safety management system.