Traffic identification is an essential stage for network management and security. As the effectiveness of portnumberbased techniques and deep packet inspection techniques is diminishing, machine learning based traffic identification has become particularly notable in the past decade. Given the importance of traffic identification, we first give a brief overview of traffic identification techniques and the basic concepts concerned, including application scenarios, input objects, identification types and evaluation metrics. Then, in the context of machine learning, we detail the development of key techniques, such as data sets acquisition, features extraction and selection, and identification model design. Additionally, we summarize and compare recent mainstream studies. Finally, we discuss the major challenges and prospects of machine learning based traffic identification.