There are primary cracks and new cracks in the engineering rock mass, and macroscopic cracks are formed, and the hollow cylindrical discrete element simulation test can realize the simulation of complex stress paths. In view of the problems existing in the hollow cylinder discrete element simulation experiment, such as the many influencing factors and the long time required for meso-parameter calibration, this paper proposes a meso-parameter calibration method for hollow cylindrical sandstone based on machine learning algorithm. By changing the different input variables in the discrete element model, 210 sets of simulation data were obtained, and a mesoscopic parameter calibration model based on random forest algorithm and XGBoost algorithm was established, the prediction accuracy of the model was compared, the sensitivity of the parameters was analyzed, and the contribution of the input parameters to the overall mechanical properties of the rock was quantified. Combined with the indoor triaxial test of the hollow cylinder, the calibration results show that the XGBoost algorithm has the advantage of calculation speed, which can quickly locate the range of discrete element mesoscopic parameters, which provides a new idea for the calibration of discrete element mesoscopic parameters of hollow cylinder, and has engineering application value.