Mobility prediction is defined as guessing the next access point(s) a mobile terminal will join so as to connect to a (wired or wireless) network. Knowing in advance where a terminal is heading for allows taking proactive measures so as to mitigate the impact of handovers and, hence, improve the network QoS. This thesis analyzes this topic from different points of view. It is divided into three parts. The first part evaluates the feasibility of mobility prediction in a real environment. It thus analyzes a mobility trace captured from a real network to measure the intrinsic entropy of the nodes motion and to measure the effectiveness of a simple prediction method. The second part investigates how to perform mobility prediction. Firstly, it examines a generic prediction scheme based on a simple machine learning method; this scheme is evaluated under various conditions. Secondly, it shows how the pieces of information that are most useful for the prediction algorithm can be obtained. The third part studies how knowing the probable next access point of a mobile terminal allows one to improve the QoS of the network considered. We deal with two situations. We first show how the handover blocking rate of a cellular network can be decreased thanks to resource reservation. We then propose a new routing protocol for delay tolerant networks (i.e. an ad hoc network where packets must be delayed in the absence of an end-to-end path) that assumes that the contacts between the nodes can be (imperfectly) predicted.