To protect public health, it is crucial to reduce the perception gap that exists between experts and the public. This can ensure appropriately informed food choices based on scientific evidence. This study aimed to build a machine learning model to predict public risk acceptance to ensure the promotion of effective food risk communication between experts and the public for bringing peace of mind to the public. We developed a machine learning model for public risk acceptance and evaluated its accuracy with risk acceptance as the output data and background information and included age, gender, health status, numeracy score, trust in public institutions, personal characteristics (self-motivation, controllability, involvement, and anxiety), favorability toward fish, risk perception, benefit perception, frequency of fish consumption, and message type as input data. We used the data from a randomized controlled study conducted in 2022 in Japan among 7,200 participants. In total, 3,324 of 3,600 men and 3,295 of 3,600 women were in the top (high-benefit and low-risk) perception category. The accuracy of the machine learning model for the prediction of risk acceptance of the public was high for both men and women with F values of 0.926 (SD: 0.000) and 0.911 (SD: 0.000), respectively. This is the first study using a machine learning approach in food safety. The accuracy of the risk-acceptance estimation model was high. The findings can help public agencies pre-design communication before disseminating health information, thereby contributing to effective benefit-risk communication via a machine learning model.