Mobile devices are becoming an integral part of our everyday digitalized life. In 2014, the number of mobile devices, connected to the internet and consuming traffic, has even exceeded the number of human beings on earth. These devices constantly interact with the network infrastructure and their activity is recorded by network operators, for monitoring and billing purposes. The resulting logs, collected as mobile traffic datasets, convey important information concerning spatio-temporal traffic dynamics, relating to large populations with millions of individuals. The thesis sheds light on the potential carried by mobile traffic datasets for future cellular networks. On one hand, we target the analysis of these datasets. We propose a usage patterns characterization framework, capable of defining meaningful categories of mobile traffic profiles and classifying network usages accordingly. On the other hand, we exploit mobile traffic datasets to evaluate two dynamic networking solutions. First, we focus on the reduction of energy consumption over typical Radio Access Networks (RAN). We introduce a power control mechanism that adapts the RAN's power configuration to the system requirements. We show that our scheme allows to significantly reduce power consumption over the network infrastructure. Second, we study the problem of topology management of future Cloud-RAN (C-RAN). We propose a mobility-driven dynamic association scheme of the C-RAN components, while taking into account users traffic demand. The introduced strategy is observed to lead to important savings in the network, in terms of signaling messages.