The precondition of realizing the scientific management of railway track is to predict the development trend of track geometric irregularity. The prediction model changes from certainty model to randomness model, which puts emphasis on the solution method for this model. In this paper, a two-stage method of support vector machine- Monte Carlo simulation (SVM-MC) for solving the prediction stochastic model of track irregularity was established. In the first stage, the ε-SVM algorithm was used to determine the model parameters which belong to small sample set, and in the second stage, the Monte Carlo simulation was used to simulate the stochastic pro- cess, and the standard deviation of the height irregularity was obtained. Compared with the previous method of track irregularity prediction, the established two-stage solution method solves the problems of small samples and non-linear prediction, and the prediction accuracy can be guaranteed when the computer capacity and speed are sufficient. A case in Shanghai and Kunming ballasted line shows that the proposed random prediction method and algorithm have good prediction results, and the average relative error is 4.63%. This method can meet the engineering application in the field and provide technical support for the maintenance plan decision.