Interdisciplinary research is fuelling a paradigm shift to endow technology-based services with a personalised dimension. The main contributors for such innovatory change are the surge in data production rate, the proliferation of data generators in the form of IoT and other network-connected devices, the incorporation of innovative data technologies like Artificial Intelligence, Machine Learning and Big Data Analytics, and the advancements in computing powers that are getting closer to dethroning Moo la and delie moe poceing pe ni ime. Moeoe, hee i an ever-increasing demand for smart and fast-responsive applications such as predictive analytics, business analysis and digital marketing. In this thesis, patientcentric cellular network optimisation is investigated as a promising paradigm that can contribute to the personalisation of present and future cellular networks with the aim of saving people lie hee ee econd con. This calls for transforming current cellular networks from merely being blind tubes that convey data, into a conscious, cognitive, and self-optimizing entity that adapts intelligently according to he e need. The work carried out in this thesis started by comprehensively exploring the role of using big data analytics in network design. Subsequently, we considered incorporating the concepts of priority, e-healthcare, Big Data Analytics, and eoce allocaion in a ingle em. The em goal i o e big daa haeed from out-patient electronic health records and body-connected medical Internet of Things sensors to be processed and analysed in a big data analytics engine to predict the likelihood of a stroke. This prediction is then used to ensure that the out-patients are assigned optimal physical resource blocks that provide good signal to interference and noise ratio (SINR) dictated by the severity of their medical state. Hence, granting channels of high spectral efficiency to the out-patients, empowering them to transmit their critical data to the designated medical facility with minimal delay. The use of several Machine Learning algorithms residing within the big data analytics engine is investigated, namely, a naïve Bayesian classifier, a decision tree
iii
classifier, and a logistic regression classifier. Further, the incorporation of the aforementioned classifiers in an ensemble system running as a soft voting classifier is examined and the performance of all classifiers is compared. The combinatorial optimisation problem of maximising he em oeall SINR hile pioiising the OPs in terms of radio resource assignment is solved using Mixed Integer Linear Programming and a heuristic. The use of two resource allocation approaches, namely, a Weighted Sum Rate Maximisation approach and a Proportional Fairness approach is considered and compared in terms of fairness and the attained SINRs. The proposed system was extended from a single-tier (homogenous) LTE-A network, to multi-tier Heterogeneous Networks employing spectrum partitioning strategy, and finally to a multi-tier Heterogeneous Network with no interferencemitigation strategies emploed. Th, enabling a fhe d of he em performance over different networks and interference strategies.
