Abstract: On the basis of the gas monitoring data of more than 40 pairs of coal mines collected in southwest China, the mathematical variation characteristics of gas concentration were analyzed, and a data-driven method and model for identifying abnormal coal mine gas emission were established. The model quantitatively evaluated the gas risk at monitoring points from different dimensions, including current gas concentration, gas mutation index, deviation from the mean value index, long-term constant index, fluctuation amplitude index and medium-term upward trend index. A single early warning index was used to identify anomalies such as trend deviation, gas mutation and so on. Based on the unit of mining face, the risk trend of regional gas disaster was quantified by means of the average of each monitoring point. Finally, a big data risk analysis platform for coal mine gas disasters was developed and applied to supervision. The application result shows that the model can accurately identify the risk of gas disasters and improves the supervision efficiency.