Abstract: There is a complex nonlinear relationship between the gas emission rate and the mining environment and mining technology. In order to accurately predict the gas emission rate, a prediction model of gas emission rate of coal seam was established by combining non-negative matrix decomposition (NMF) and random forest (RF). The dimension reduction of 13 characteristic indexes of the data from a mine in Shenyang was reduced by non-negative matrix decomposition.The extracted 5 main components were used as the inputs of RF, and the gas emission rate was used as the output to realize regression prediction. Hunger Games Search Algorithm (HGS) was used to optimize the parameters of RF, and the results were compared with those of Sparrow Search Algorithm (SSA), Coronavirus Herd Immunity Optimization (CHIO), Genetic Algorithm (GA), Arithmetic Optimization Algorithm (AOA), Aquila Optimizer (AO) and Archimedes Algorithm (ArchOA). The experimental results show that the convergence speed of NMF-HGS-RF is relatively fast and the prediction accuracy is relatively high. The average relative error of NMF-HGS-RF is 4.74%. The average relative errors of NMF-RF, NMF-SSA-RF, NMF-CHIO-RF and other models are 6.34%, 5.85%, 6.21%, 8.49%, 4.95%, 6.11% and 7.93% respectively. The results show that the prediction accuracy of NMF-RF is higher than that of RF without dimension reduction, and NMF-HGS-RF has better prediction performance than RF after optimization of other parameters.