Objective Conducting food safety monitoring and early warning for merchants is a critical issue in China?? s food safety supervision. This study aims to achieve efficient food safety early warning and supervision. Methods In combination with the characteristics of food safety in Chongqing comprehensive evaluation indicators were obtained through expert assessment. An AHP-BP neural network comprehensive evaluation method was proposed and a reliable and objective food safety evaluation system was established. Based on this dynamic mining of food safety feature indices was conducted to generate a comprehensive food safety profile. A food safety early warning and monitoring model based on random forest was established and the model was verified through simulation using food safety data from Jiangbei District Chongqing. Results The comprehensive evaluation indicator system is more rational with significant improvements in prediction accuracy and runtime compared with the XGBoost algorithm. Conclusion Machine learning methods contribute to the establishment of a more comprehensive and rational food safety evaluation system. The food safety early warning model based on random forest performs better in terms of precision AUC and recall rate. It not only achieves high accuracy in food safety supervision for merchants but also demonstrates good robustness.