In seismological research, picking speed and accuracy of seismic detection and phase identification directly affect their application efficiency and accuracy in precise seismic positioning and tomography. In recent years, machine learning has attracted wide attention in the field of seismology. Machine learning can improve upon traditional seismic detection and phase identification methods, thus achieving more accurate and higher recognition rates. In this paper, we introduced a machine learning method according to the classification of supervised learning and unsupervised learning, then summarized the flow of the machine learning method. Finally, we reviewed those machine learning methods widely used in seismic detection and phase identification, i.e., convolution neural network, fingerprint and similarity threshold, generalized phase detection, PhaseNet, and fuzzy clustering. Results showed that machine learning will be the primary means of seismic event detection and seismic phase identification. Application of data-driven machine learning in seismology combined with the physical model will be the development trend of the future.