As the major crushing equipment for the pretreatment of coarse ore in mining, the productivity of the gyratory rotary crusher is affected by its structural parameters, working parameters, feeding amount and other factors. Based on the DEM and Ab-T10 particle crushing model, an equal ratio 62–75 gyratory crusher model was established. The effects of rotating speed of mantle shaft, feed particle size, open side setting size and rock strength on the working performance of gyratory crusher were numerically studied, and three tree-based machine learning methods were used to predict the productivity of gyratory crusher. The results show that the feature importance of ore strength under the three regression models is the largest, and the prediction performance of Extra Random Tree (ET) is the best, followed by AdaBoost regression model, and the prediction performance of Random Forest (RF) is slightly worse. It is recommended to use the ET to predict the productivity of the gyratory crusher. The results obtained of this paper provide some reference and guidance for optimizing the structure of the gyratory crusher and selecting the type according to different working conditions.