Primary cracks and new cracks develop within the engineering rock mass, leading to the formation of macroscopic cracks. The hollow cylindrical discrete element simulation test enables the emulation of complex stress paths. In order to solve the problems existing in the simulation test of hollow cylindrical discrete element, such as numerous influencing factors and lengthy mesoparameter calibration, a method of mesoscale parameter calibration of hollow cylindrical sandstone discrete element based on machine learning algorithm was proposed. Through variations in input variables within the discrete element model, 210 sets of simulation data were obtained. A mesoscopic parameter calibration model based on random forest algorithm and extreme gradient boosting(XGBoost) algorithm was established, the prediction accuracy of the model was compared, the parameter sensitivity was analyzed, and the contribution of input parameters to the overall mechanical properties of the rock was quantified. Combined with the indoor triaxial test of hollow cylinder, the calibration results show that the XGBoost algorithm has the advantages of computing speed, and 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 the value of engineering application.