Abstract: To predict deformation and locate damaged areas of roadways, a numerical model for the surrounding rock in a roadway with multiple disturbances on the floor was established. A dataset of roof deformation of the bottom drainage roadway with different geological parameters, such as surrounding rock strength and lateral pressure coefficient, and mining parameters, such as roadway section, drainage borehole and support strength was obtained. Machine learning algorithms, such as random forest, extremely randomized trees, GBDT and XGBoost were used to establish single-based learner roadway deformation prediction models respectively. With the elastic net algorithm as the meta-learner and using the Stacking fusion method, the output models of different based learners were fused to construct a fusion prediction model for surrounding rock deformation of the bottom drainage roadway under multiple disturbances.The inhibiting or promoting effects of various characteristic factors on roadway deformation were evaluated, and the dominant controlling factors affecting the stability of the surrounding rock in the bottom drainage roadway were identified. The bottom drainage roadway of the transportation roadway No. 14040 in Zhaogu No. 2 Mine was chosen as the engineering background. Using the established roadway deformation prediction model, with the actual production geological conditions and mining parameters of the roadway as input , the recommended support strength for the roadway was determined through reverse calculation by setting the desired roadway deformation, which guided the on-site roadway support design and key parameter determination. After field implementation of the recommended support strength, the roadway deformaitn is controlled within the allowable deformation range as specified by the decision-making model. The roof deformation is only 52% of the original support, effectively controlling the large deformation of surrounding rock in roadways. The roadway deformation prediction model established based on machine learning provides a new approach for roadway stability maintenance, promoting the development of intelligent operation and maintenance technology for coal mine roadways.