Today's data networks are complex entities that operate in dynamic and heterogeneous environments. The architecture and protocols of these networks have to face several challenges such as dynamic adaptation and autonomous decision-making in the presence of several, often conflicting, criteria such as delay, loss rate, jitter, energy, etc. However, multicriteria decision making problems usually have multiple solutions. These problems are solved by methods that use parameters whose choices have consequences difficult to predict. Thus, most multicriteria decision making methods proposed in the literature assume the presence of a decision maker who guides the decision process. Finally, the choice of parameter values often involves an interaction with the decision maker, which is difficult or impossible to do in an autonomous context. In this thesis, we propose a new multicriteria decision making method suitable for autonomous systems in general and autonomous networks in particular. The Russian doll like method we propose uses a set of nested quality boxes (like Russian dolls) defined in the criteria space, in order to approximate a wide range of utility functions. First, the proposed method adapts to the dynamic nature of autonomous networks in order to maximize user satisfaction. Second, it uses parameters that are directly deduced from objective facts such as technical standards or specifications, or obtained from a MOS (Mean Opinion Score) experiment using an automatic classification method. We tested the performance of the Russian doll like method in a case of routing in wireless ad hoc networks. Experiments have shown that routing done with the Russian doll like method is always better or similar to the one done by the weighted sum method which is widely used. This is due to the adaptation ability of the decision provided by this new multicriteria decision making method
