Abstract: Food safety risk early warning is a crucial technical approach to ensure that regulation stays ahead of potential risks. In recent years, machine learning, as an emerging technology, has demonstrated its immense potential in food safety risk early warning due to its powerful data processing and analysis capabilities. This paper introduced the concept of food safety risk early warning and the current food safety risk early warning systems both domestically and internationally. It reviewed the main principles, basic characteristics, and application progress of various machine learning methods, including logistic regression, least absolute shrinkage and selection operator, support vector machine, random forest, Bayesian network, extreme gradient boosting, light gradient boosting machine, and artificial neural networks. The paper also presented the advantages and disadvantages of machine learning in current food safety risk early warning application scenarios, as well as future development directions. In particular, advanced technologies such as multimodal data fusion and deep learning are expected to play an increasingly important role in future food safety risk early warning.