Abstract: In current gas concentration prediction methods, there are problems of data processing uncertainty, feature extraction limitations, and prediction bias caused by subjective factors. In order to solve the above problems, a gas concentration prediction method for coal mine excavation working face is proposed. Firstly, laser gas sensors are installed every 1 meter in the return airway of the coal mine excavation working face, forming a sensor network to collect real-time gas concentration data. Secondly, the method searches and removes outliers in the gas concentration data according to the Laida criterion, and uses the Lagrange interpolation polynomial to fill in the missing values in the gas concentration data. Finally, based on removing outliers and filling in missing values in the gas concentration data, the empirical mode decomposition algorithm is used to decompose the gas concentration data into intrinsic mode functions and trend terms. The Hilbert transform is then used to process the intrinsic mode functions to obtain the high-frequency and low-frequency terms of the data, which are then input into the least squares support vector machine for weighted processing to output the gas concentration prediction results. The gas concentration prediction simulation experiment is conducted using a simulation device for the excavation working face, and an on-site test is conducted on a certain coal mine excavation working face. The results show that the predicted gas concentration by this method is very close to the actual measurement value, with a small mean square error, indicating a high accuracy of the prediction results. The small fluctuation of mean square error indicates good adaptability and strong stability of prediction results. Short prediction time indicates high prediction efficiency.