pour obtenir le grade de

Modern technologies allow to access to more and more detailed information on human interactions. In this context, the SocioPatterns collaboration has allowed to develop an infrastructure based on radio-identi cation devices, that records human proximity patterns at a ne grained resolution, among voluntary individuals. This infrastructure has been deployed in diverse contexts, such as scienti c conferences, a museum, a primary school, or a hospital department. The mere analysis of these data represents a high stake for the study of human dynamics and raises fundamental issues such as the need of adequate tools and analysis techniques. This thesis presents the statistical characterization of physical proximity dynamics, put into relation with the context and other available metadata such as the age, the gender of participants or the structure of their virtual social networks. Although contact patterns considerably di er amongst the various contexts, the empirical distributions of interaction durations and of inter-contact times are very similar. An agent-based model, presented in this thesis, suggests simple microscopic interaction rules able to produce the complex macrostructure of interaction durations. In the last place, the characterization of contact dynamics constitutes a determining step for understanding spreading mechanisms of diseases such as the in uenza. The human proximity data have allowed to analyze the level of information needed on contact dynamics for the elaboration of epidemiological models of contagion. Such models allow to better estimate the impact of public health strategies, e.g. the closure of school classes and targeted vaccinations.tel-00777540,version1-