Surface-enhanced Raman spectroscopy (SERS) offers significant advantages in the on-site rapid screening of food-safety risk factors due to its high sensitivity, specificity, and cost-effectiveness. However, the widespread application of this technique still faces challenges, including high-dimensional spectral data handling, interference from complex food matrices in trace-level detection, and difficulties in resolving overlapping spectral peaks. Recent advances in deep learning (DL) and machine learning (ML) have provided innovative solutions for SERS data analysis. The integration of ML methods (especially multivariate tools) with SERS enables efficient processing of complex spectral data, significantly improving detection performance, and has become a research hotspot. This review first briefly introduces the fundamentals of SERS and ML. Next, it highlights the application of SERS combined with ML (SERS-ML) in detecting food safety risk factors, such as pathogens (e.g., bacteria, viruses), organic/inorganic toxins (e.g., pesticides, antibiotics), and microplastics (MPs), with an emphasis on their identification and quantification. Furthermore, the key challenges and factors for the application of SERS-ML to complex food systems are discussed. Finally, the practical application potential of SERS-ML integration is outlined to inspire further research and technological innovation.