The 4R training method is used to train workers in many industrial workplaces, to reduce accidents caused by human factors. The 4R training method enables trainees to develop their hazard prediction abilities and response capabilities to avoid hazardous situations. In general, this training involves identifying hazards shown in hazard prediction training sheets. However, there is a significant problem with the 4R method: a single trainee cannot train themselves using the 4R method, because this training requires the assistance of an expert instructor. To solve this problem, this study aims to develop a hazard prediction training system. This system enables trainees to use the 4R method to train themselves anytime and anywhere. This paper provides a summary of the proposed training system, which uses a machine learning method to generate a subsystem for evaluating the trainee answers. Experimental results show the accuracy rates (precision) for two sheets were 64 ± 14 [%] and 70 ± 12 [%].