Distributed Brillouin fiber sensors can measure the temperature and strain information along optical fibers over distances of hundreds of kilometers. They are used to monitor major national projects such as bridges, tunnels, power lines, and oil and gas pipelines. The main method of Brillouin sensing is to measure the Brillouin frequency shift, which is linearly related to the temperature and strain of an optical fiber. The Brillouin frequency shift is usually obtained by measuring the Brillouin spectrum of the optical fiber. The spectral line of Brillouin spectrum theoretically satisfies a Lorentz line shape, and the frequency corresponding to its peak is Brillouin frequency shift. To reduce the influence of sampling accuracy and noise, the most common method used to extract the Brillouin frequency shift from the Brillouin spectrum is Lorentz curve fitting. However, curve fitting is sensitive to initial values and the fitting error significantly increases when the signal-to-noise ratio is low. In addition, the processing time of curve fitting is too long, which reduces the response speed of the system. To improve the accuracy and speed of Brillouin frequency shift extraction, machine learning has recently been applied in this field, which has provided better results than traditional curve-fitting algorithms. This article mainly presents the achievements of machine learning in Brillouin frequency shift extraction in recent years, including singular value decomposition, support vector machines, and artificial neural network technology.