Abstract: In order to establish a high-precision prediction model of mine slope displacement, Doubly Sparse Relevance Vector Machine （DSRVM） based on Particle Swarm Optimization （PSO） was used to establish the nonlinear relationship between slope stability and influencing factors in this paper. DSRVM was a multi-core combinatorial optimization method, which was proposed under the framework of variational and Relevance Vector Machines （RVM）. Compared with RVM and other multiple-kernel learning methods, DSRVM not only had less training time, but also can obtained higher prediction accuracy. Aiming at the influence of the parameter’s selection of DSRVM on the final prediction effect, the optimal multiple kernel parameters was determined by PSO algorithm to be used in the mine slope displacement prediction. Compared the computational results of DSRVM with Extreme Learning Machine （ELM） and Wavelet Neural Network （WNN）, the feasibility of PSO-DSRVM in slope deformation prediction was verified by the evaluation indicators such as RMSE, R² and ARPE.