Continuous microseismic monitoring was carried out around 9 producing longwall top coal caving (LTCC) panels with concurrently recorded daily face advance rates at Coal Mine Velenje in Slovenia over a 27-month monitoring period. The monitoring results suggested that spatial and magnitude characteristics of microseismicity are dominated by those of underlying fractures, while microseismic event rate is under the combined effects of local natural fracture abundance and mining intensity. On this basis, a data-driven yet physics-based forecasting methodology was established for LTCC induced hazardous microseismicity, which is above a given threshold of energy magnitude and within a certain distance to the longwall face. Statistical analyses were first conducted to characterise temporal, magnitude and spatial characteristics of long-term recorded microseismicity, based on which a short-term forecasting model was developed to calculate the probability of hazardous microseismicity considering the three characteristics. The model developed was employed to forecast the likelihood of hazardous microseismicity at one of these LTCC panels, and the forecasted results were supported by the monitoring. This statistical model has important implications in the evaluation of mining-induced hazards, and it can be used to optimise longwall face advance rates to minimise the risk of hazardous microseismicity in burst-prone deep-level mining sites.