<p>The challenge of obtaining large-scale geostress greatly limits its widespread application in predicting coal-and-gas outbursts. This paper presents a coal-and-gas outburst risk classification prediction method based on the evolution of the entire space–time stress field in a mine. The approach involves three levels of risk prediction: a primary assessment using in situ stress data and a rock damage risk coefficient formula, a secondary refinement using a neural network combined with the entire space–time stress data, and a tertiary prediction based on mining-induced stress forecasts. Applied to the North 1 mining area of the Sangshuping coal mine, the method successfully divided three levels of risk zones, with 73.4% of recorded outbursts occurring within the high-risk zones identified in the primary assessment. The second-level refined prediction model achieved strong accuracy (0.97), and the proportion of recorded outbursts in high-risk zones increased to 88.68%. Simulations of future mining activities revealed an obvious high-risk zone in the 4321 and 4322 working faces. The study highlights the relationship between stress fluctuations and risk severity. This method is of great significance for accurately predicting coal-and-gas outbursts and guiding on-site safety production in coal mines.</p>