Due to the complexity of in-situ experiments and scale limitations for ground-based facilities, the numerical simulation of hypersonic flows is currently one of the key research methods in gas dynamics. For hypersonic flows due to high temperatures behind the shock wave, it is necessary to take into account the physical and chemical processes in order to simulate the flow correctly. The most accurate calculations of transport coefficients for air components can be made with the kinetic theory of transport and relaxation processes, using the Chapman-Enskog method for solving Boltzmann equations. However, such an approach leads to the necessity of extensive calculations of collision integrals at each step of the iteration procedure, the number of which in the system increases with increasing accuracy of the solution. On the other hand, there are a number of approximate formulas for calculating transport coefficients, such as Blottner model for viscosity and Eucken relation for thermal conductivity. Applying the regression algorithms (including neural networks) is promising in terms of balancing speed and computational accuracy. The hypersonic flow of multicomponent air past a sphere is used as a model problem, employing OpenFOAM - the open-source platform for numerical simulation of continuum mechanics problems - for calculating flow parameters. Various approaches to the application of machine learning methods are investigated.