A survey on machine learning methods for food safety risk assessment: Approaches, challenges, and future outlook

Abstract

Food safety is essential for protecting health and supply chain management. Food engineering plays a foundational role in ensuring food safety by developing innovative processes and technologies for quality control, contamination monitoring, and risk mitigation. Risk assessment is an effective means to ensure food safety, while machine learning (ML) is crucial in facilitating this process. It improves the accuracy of food quality inspection and the speed of risk assessments through rapid learning and processing of data. This survey provides a comprehensive analysis of commonly used supervised and unsupervised learning methods for food safety risk assessment, highlighting advancements, challenges, and future directions. Supervised learning methods, such as Bayesian network (BN), support vector machine (SVM), artificial neural network (ANN), etc., are successfully applied in prediction of food safety risks and improves the prediction accuracy and efficiency. Unsupervised learning methods, such as k-means, hierarchical cluster analysis (HCA), autoencoder (AE), etc., perform well for unlabeled or high-dimensional food anomaly data. The review also addresses key challenges in the food field, such as class imbalance, the emergence of new and unexpected risks, and the integration of multi-source heterogeneous data, including regulatory data, e-commerce sentiment, and public opinion. The utilization of large language models (LLMs), few-shot learning (FSL), and knowledge graphs together offers promising solutions to key challenges in food safety risk assessment. This comprehensive survey emphasizes the transformative potential of ML in enhancing the field of food safety risk assessment and management.