The next generation 5G wireless networks have extreme requirements for ultra-high availability and performance, demanding quality of service (QoS), and very short delay for services. Furthermore, applications such as live streaming, videos, e-commerce, e-Health, autonomous systems, etc., require a massive amount of bandwidth and become much more complex. This is coupled with complex ICT system structure (Devices, Software, and Protocols) with different interaction players (Customers, Service Providers, Network Operators, etc.). In addition, unusual temporal fluctuations in cellular network traffic could lead to drastic network management misbehaviors and drop in quality of experience (QoE).
Since there is a technological limit to what network devices can achieve in terms of availability and performance, another solution to improve efficiency is to investigate the behaviors of users from the perspective of data-driven approaches using the input and output of the system’s behavior. This approach will lead to effective user behavior prediction. The predicted user behavior can then be used to improve service delivery in terms of availability, resource management, performance and QoS/QoE. One approach to achieving this objective is to machine learning (ML) techniques coupled with an effective data-driven approach.
The aim of this research is therefore to investigate effective data-driven ML techniques for 5G networks. Two main tasks are to: (i) collect, measure, process, and store data traffic and log elements, and (ii) use simulation and testbed to identify the most effective set of ML algorithms suitable for analyzing and predicting both 5G system’s behaviors and user behaviors for performance improvement. The key motivation is that future management of 5G networks will heavily rely on data-driven ML techniques due to the huge volume, diverse varieties of applications, and fast velocity of traffic.