SummaryThe excess of gas concentration in the top corner of coal working face has always been the main factor restricting the safe productivity of coal mines. Therefore, the rapid and accurate prediction of top corner gas concentration is an effective method to prevent gas disasters. At the same time, the development of the Internet of things has made the gas monitoring data collected by the coal mine safety monitoring system exhibit nonlinear big data characteristics. In order to mine the characteristic data related to the gas concentration of the top corner from a high‐dimensional and nonlinear monitoring datasets, a model that integrates the t‐distributed Stochastic Neighbor Embedding algorithm (t‐SNE) and the Support Vector Regression (SVR) algorithm to predict the gas concentration of the top corner on the coal working face is proposed. First, the multidimensional monitoring data are nonlinearly dimension‐reduced by t‐SNE algorithm, which enabled the spatial feature data of the monitoring data to be extracted. After that, the SVR algorithm was used to construct the nonlinear regression model between the spatial feature data and the actual gas concentration of the top corner to predict the gas concentration of the top corner. The experimental results show that the predictive model based on t‐SNE and SVR was better than the multiple linear regression, SVR, Principal Components Analysis (PCA) + SVR. The results show the model based on t‐SNE and SVR was more stable and could provide more accurate predictions, anomaly sensitivity, and the fitness is 0.55628405, which can better fit the actual gas concentration of the top corner.