This work considers the application of machine learning methods for approximate determination of diffusion coefficients that are part of extended Navier–Stokes equations solved in a state-by-state approximation. Three methods are suggested: the k-Nearest Neighbors (k-NN) algorithm, a classical neural network (NN) and Physics-Informed Neural Network (PINN). The resulting solution, fully based on data and well-known physics relations, can be used to direct research in more complex cases.