Profiling and identifying users' activities with network traffic analysis

Traffic identification technique is used for classification of different network protocols and applications even with detection of users' network activities. In this paper, we conduct our study on some typical users' network activities and present a traffic identification method to describe the feature about users' behaviors. We convert users' network activities information into different sequences with frame size and inter-arrival time just like HMM model does. Then we use latent dirichlet allocation model to extract potential characterization from these activity sequences. We use some machine learning algorithms to classify different activities in the data, and we suggest a method to evaluate the mixed activities within a sequence. Our research shows that it is possible to distinguish exact users' on-line activities without inspection to packet payloads, and users' privacy may need more protection against traffic analysis.