Abstract: The spatiotemporal distribution characteristics of microseismic events are closely related to the intensity of rockbursts in coal mines. Prediction methods and visual representations of microseismic data based on machine learning are important technical means for monitoring and early warning of rockbursts. However, challenges remain in effectively predicting the movement paths of hazard areas, visualizing spatiotemporal patterns of microseismic data in a coordinated manner, and clearly displaying stacked data. To address this issue, a prediction method for rockburst hazard areas based on a microseismic spatiotemporal flow map was proposed. In the data preprocessing module of this method, a two-dimensional kernel density estimation method was used to continuously represent discrete microseismic data, and a kernel density heat map was constructed to reflect the spatial aggregation degree of microseismic events. In the spatiotemporal flow map construction module, the gravity model was improved to extract the spatiotemporal features of microseismic data, and arrows were used to visualize the direction of the spatiotemporal flow. In the hazard area prediction module, the K-means clustering algorithm was used to optimize the visualization results. Microseismic data from the 2215 working face of a mine in Inner Mongolia and the 4106 working face of a mine in Shaanxi were used to conduct rockburst hazard prediction experiments separately for fault zones and goaf areas. The results showed that this method could effectively and accurately predict the movement direction of rockburst hazard areas. After applying the K-means clustering algorithm, the number of arrows in the microseismic spatiotemporal flow maps was optimized by 77.27% and 87.5%, respectively, making the visualizations more concise and intuitive.