Brillouin optical time domain analysis (BOTDA) has shown its unique advantages in distributed optical fiber sensing systems and has received widespread attention. The rapid and accurate extraction of temperature distribution information in BOTDA sensing system is extremely desirable. With the rapid development of machine learning, it shows great potential in temperature extraction of BOTDA sensing system. First, the principle of BOTDA sensing system is introduced. Then, some machine learning algorithms are illustrated and their applications and advantages for temperature extraction of BOTDA sensing system are analyzed. Finally, outlook for future research is given.