AP and MN-centric Mobility Prediction: A Comparative Study Based On Wireless Traces

The mobility prediction problem is defined as guessing a mobile node's next access point as it moves through a wireless network. Those predictions help take proactive measures in order to guarantee a given quality of service.
Prediction agents can be divided into two main categories: agents related to a specific terminal (responsible for anticipating its own movements) and those related to an access point (which predict the next access point of all the mobiles connected through it).
This paper aims at comparing those two schemes using real traces of a large WiFi network. Several observations are made, such as the difficulties encountered to get a reliable trace of mobiles motion, the unexpectedly small difference between both methods in terms of accuracy, and the inadequacy of commonly admitted hypotheses (such as the different motion behaviours between the week-end and the rest of the week).