Mathematical models for complex systems are often accompanied with uncertainties. The goal of this paper is to extract a stochastic differential equation governing model with observation on stationary probability distributions. We develop a neural network framework to learn the drift and diffusion terms of the stochastic differential equation. We introduce a new loss function containing the Hellinger distance between the observation data and the learned stationary probability density function. After minimizing this loss function in the training procedure, we find that the learned stochastic differential equation shows a reasonable approximation of the data-driven dynamical system. The effectiveness of our framework is demonstrated in numerical experiments.