Abstract: In order to avoid the problems of the traditional prediction methods in the prediction of the support anti-destruction ability of the high stress soft rock roadway, a prediction method of the support anti-destruction ability of the high stress soft rock roadway based on machine learning is proposed. Firstly, according to the physical and mechanical properties of the rock, the strength characteristics of the soft rock are analyzed. At the same time, under the support of support construction model, the rheological characteristics of surrounding rock of soft rock roadway are analyzed, and the characteristics of high stress soft rock roadway are studied. Then, according to the plastic deformation mechanism of soft rock, the corresponding relationship among elastic zone, plastic hardening zone, plastic softening zone and plastic flow zone is determined, and the resultant force of surrounding rock movement to the free area is calculated to determine the optimal support time. The machine learning method simplifies the support structure of high stress soft rock roadway into a gourd structure model, calculates the external force, and determines the support structure quality parameter matrix of high stress soft rock roadway. By comparing the three conventional methods, it can be seen that the prediction accuracy of this method is higher under three kinds of different failure strength.