In order to identify the key factors inducing the surface subsidence while shield tunneling，a key parameter selection model and solving method based on RS-SVM is proposed. The information entropy rules were used to discretize seven continuous variables including Internal Friction Angle(IFA)，Cohesive Force(CF)etc. Genetic Algorithm and Rough Set for attribute reduction were combined to obtain several collections that significantly affect the surface subsidence; Using the statistical learning of RVM to select the best collection which optimally reflects the relationship between parameters and surface subsidence，we get four parameters: Single Ring Grouting Pressure，Internal Friction Angle，Specific Torque(ST)，Incision of Slurry Pressure and each of the collections was the critical parameter that should be considered in construction. The method was applied in a completed metro tunnel in Wuhan，China and the results indicated the feasibility of the method.