Abstract: This paper conducts a study on the issue of human unsafe behavior and constructs miners’ unsafe behavior indicators system from three aspects of design and use, management and behavioral mistakes. Combined with the expert consultation and the actual situation of coal mine, this paper tries to use the genetic algorithm to optimize neural network and constructs a coal mine safety prediction and evaluation model based on miners’ unsafe behavior. Sample learning and empirical analysis is made based on real-time monitoring data collected from large-scale coal mines in Shandong, Henan and some other places. The results show that a short-term prediction results is consistent with the actual situation and able to predict the coal mine safety situation in advance. It is feasible to use genetic neural network algorithm to predict the overall safety of coal mine in real time and the feedback of results may provide decision support for managers for sure.