Abstract: In order to improve the prediction precision and accuracy of oil-type gas content in mines, we propose an accurate determination method based on multivariate data fusion. Based on the main factors affecting the oil-type gas content, and 27 sets of actual measurement data from the mine were collected, and the XGBoost algorithm was used to screen out buried depth, roof and floor lithology, fold and porosity as the key features which were standardized to ensure that the data with different magnitudes could be reasonably fused in the modeling process. Four classical machine learning algorithms, namely, Kriging interpolation, least squares support vector machine, multilayer perception and gradient boosted regression tree, were used for preliminary prediction, and comparative analysis was carried out for the regression problem of oil-type gas content. The results show that the gradient boosting regression tree algorithm performs best in terms of prediction performance, with a coefficient of determination of 0.987, a normalized mean square error between 0.001 and 0.010, and a total information criterion between 0.019 and 0.046. The prediction accuracy is further improved by combining the Stacking algorithm. The Stacking method fuses the prediction results of multiple base learners as new features by using an improved whale optimization algorithm to optimize the weights of each base learner. In order to further improve the prediction ability of the model, a bidirectional long and short-term memory network is introduced, and the final fusion model is constructed through a meta-learning mechanism to deeply learn the prediction results of the base learners in order to capture more complex nonlinear relationships and temporal information. The fusion model significantly outperforms the traditional single algorithm on the test set. The average absolute error of model prediction is 0.116 m³/t, the average value of the normalized mean square error is 0.006, the average value of the total information criterion is 0.004, and the coefficient of determination is higher than 0.98, which shows its high accuracy and stability in the prediction of oil-type gas content in mines.