Enhancing QoS in 5G networks using Self Optimization of Radio Resource Management Parameters The demand for high data rate mobile traffic is increasing tremendously as the world transcends into High Definition (HD) quality applications, video calling, streaming traffic, social media etc. To match these sky-rocketing user demands, increasing traffic and volatile radio environment, mobile networks are continually evolving and becoming more and more sophisticated. While, the trend of mobile networks has been towards an all IP flat network, the network Quality of Service (QoS) metric has shifted from simple voice services to providing high volume data services. The increased network complexity puts a high burden on operation and maintenance costs making the traditional methods obsolete. In this backdrop, the concept of Self Organizing Networks (SONs) was introduced in the 4G mobile network standard by the 3rd Generation Partnership Project (3GPP) to enhance network performance and reduce operational costs. SON is also a significant component in the upcoming 5G mobile standard and thus has received much interest by the research community. SONs behave like an intelligent living organism and adapt to changing environment, resources and traffic loads. Two areas that have a notable impact on network performance are, interference mitigation and coverage adaptation for load balancing and these are the main focus of this PhD research work. We have worked on finding and comparing different self-optimisation techniques based on network Key Performance Indicators (KPIs), to reduce network interference and balance traffic load in the context of SON. In particular, we have applied simple machine learning techniques of Stochastic Cellular Learning Automata (SCLA), simple Q-Learning and Artificial Neural Networks (ANN) QLearning in a fully distributed SON 5G environment with a unique information sharing model among cells, its neighbours and the network. This model is unique in the sense that it depends on a simple distance separation criteria instead of Radio Frequency (RF) environment to identify and define neighbours for information sharing. Interference reduction was done for femtocells, and coverage adaptation for load balancing was done using active antenna tilt model. Test results from network-based simulators based on 3GPP guidelines show that simple SON technique like SCLA adapt quickly, as compared to advance techniques like Q-Learning but are limited in capturing complex network scenarios. The reason being, simple Q-Learning techniques fail to swiftly adjust to changing environment conditions as the number of state variables grow. This is due to increased training time required to build a meaningful Q matrix. ANN showed promising results concerning agility and adaptability to complex changing environments. ANN has the inherent capacity to accept a large number of inputs, reduce the input dimension and adapt to changes as time grows. It is thus concluded, that simple machine learning techniques like SCLA are best suited for enhancing QoS in 5G networks where optimisation input variables are unavailable or unknown like in standalone Femtocell case. However, in scenarios where the numbers of input variable are known and readily available from the network, i.e. cooperative distributed environment, ANN gives better results.