Mobile operators are continuously challenged to offer faster and better connectivity to their customers. The mobile networks are expected to provide enhanced capacities while dealing with a new variety of services. 5G brings a wide range of novelty to increase the quality and services offered by the operators. These novelties include, but are not limited to, new antenna technologies, the apparition of Multi-User scheduling, new architectures, but also the inclusion of more artificial intelligence in the network decisions. The changes brought by 5G vastly increases the complexity of the networks. 5G comes with a wide range of new parameter to be set in an optimized manner. The demand towards the operators are higher than ever whereas the network is more complex. In 4G, the number of parameters to be chosen was already high and complex. To optimize the network and enable the users to get the best quality of experience, Self-Organizing Network (SON) had been introduced in 4G network. SON allows the network to enable self-optimization, self-configuration and self-healing. The SON has to evolve to deal with the new challenges of the network. M-MIMO and beamforming present the opportunity to focus signal on users, and thus increasing throughputs. However, focused signal come with new complexity and new challenges to overcome. We consider in this thesis the self-optimization of network relying on M-MIMO with Multi-User (MU)-schedulers. First, we consider a heterogeneous network. Densifying a network by adding small-cells increases quickly the capacity of the network. Massive Multiple Input Multiple Output (M-MIMO) and MU-scheduler render the 4G SON solutions out-of-date and present a new challenge to overcome. In this thesis, we propose a MU-collaborative scheduler for 5th Generation (5G) M-MIMO cell that takes into account the impact of the macro scheduling on the small-cells located in the macro cell’s area. We also assess the performance of the scheduler through numerical experiments. Interferences between neighboring M-MIMO cells are also an important topic to be addressed in 5G network. In this work, we propose three different approaches to this problem. The Automatic Neighbor Beam Relation (ANBR) solution is a heuristic solution offering a low complexity and a reactive solution to deal with neighboring cells. The thesis then introduce a Multi-Armed-Bandit (MAB) solution to enhance the previous heuristic to deal with neighboring interferences in a 5G M-MIMO network with online learning. Finally, a third approach based on the exploitation of geolocalized data to take advantage of information at a finer granularity is explored.