Abstarct: Aabstract Support vector machines are powerful tools for data classification and regression. In recent years, many fast algorithms have been developed for support vector machines. Support vector machines are used in many military and engineering applications, real time issues, such as classification in complex electromagnetic environments, diagnostics in medicine, and so on. Given that numerical convergence methods require more calculations and can not meet our needs in real-time problems. One promising process for training real-time support vector machines is the use of recurrent neural networks baxsed on periodic execution. In this thesis, we introduce two neural networks to solve support vector problems such as support vector regression problem and stochastic support vector regression with probabilistic constraints . We prove that the equilibrium points of the neural networks are equivalent to the optimal solution of the support vector problems. The stability and convergence of the proposed neural networks will also be demonstrated. We confirm the theoretical results by providing examples.