Abstract: To address the issues of insufficient prediction accuracy and slow convergence of single characteristic gas concentration, this study proposed a gas concentration prediction model based on the hybrid approach of STL-CEEMDAN-ISSA-LSTM. Initially, the gas concentration was decomposed into the trend, periodic, and irregular terms using the STL technology. To address the randomness and uncertainty associated with the irregular term, CEEMDAN decomposition was employed to separate it into 11 IMF components and residual margin. To improve the searching ability of the Sparrow search algorithm and to optimize LSTM-related hyperparameters, this study introduced the Sin chaos model, reverse learning strategy, and Cauchy mutation strategy. The ISSA-LSTM gas concentration prediction model was constructed to predict the trend term, periodic term, IMF components, and residual margin decomposed by CEEMDAN. The prediction results of each component model were then superimposed to obtain the final prediction results of gas concentration. Finally, the proposed model was compared with LSTM, STL-CEEMDAN-LSTM, STL-ISSA-LSTM, and other models using the measured gas concentration in Yanbei Coal Mine 250203 working face as a sample. The results demonstrate the superior prediction accuracy of the proposed model over other algorithms.