Development and utilisation of predictive modelling tools for optimising the operations of future cellular networks

Rizwan, Ali (2025) Development and utilisation of predictive modelling tools for optimising the operations of future cellular networks. PhD thesis, University of Glasgow.

Abstract

With the advent of ultra-densification and the proliferation of diverse underlying technologies, 5G and beyond networks promise unprecedented capabilities but also introduce significant challenges in network management due to the manifold increase in complexity. Two such key challenges are network resources optimization and network maintenance. It is challenging to perform these tasks manually, with 5G and beyond networks it does not remain viable at all. Self-aware solutions like Self-Organizing Networks (SON) have, therefore, been proposed and well explored in research to address these challenges.

However, the legacy SON, still relies on the predefined instruction sets making them inelastic to network changes. On the other hand, they use the results from field tests or customers complaints for the identification of network issues which cause unnecessary delays and make existing SON reactive. But, to cater for the exponential increase in network complexity and diverse use cases in 5G and beyond networks, SON needs to be more intelligent, adaptable and autonomous. They are required to be proactive rather than reactive. Artificial intelligence can play a crucial role here and machine learning can equip SON with this intelligence by exploiting the hidden patterns in the real network data. Another advantage of machine learning is that it also brings prediction capacity important for making SON proactive.

Artificial intelligence, particularly machine learning, offers the potential to transform SON into proactive systems by extracting actionable insights from network data and enabling predictive capabilities. This research focuses on leveraging machine learning to enhance two key SON functions: self-optimization and self-healing.

This study begins by exploring and classifying various data types generated within the network, highlighting their potential roles in wireless cellular networks (WCN). A review of existing and potential use cases for SON functions and machine learningbased approaches provides the foundation for this work. For self-optimization, a Support Vector Machine (SVM) model is developed to predict internet traffic loads using Call Detail Records (CDR), achieving up to 91% prediction accuracy. Additionally, a novel machine learning model leveraging Geohash global indexing predicts users’ next locations with approximately 95% accuracy, marking a significant contribution to mobility prediction.

To enable self-healing, a hybrid machine learning scheme is proposed. Using CDR data, network cells are grouped based on performance via K-means clustering. Subsequently, an SVM classifier is employed to categorize cell performance with 98% accuracy. The traffic prediction model for self-optimization is further refined through a Support Vector Regression (SVR) approach, achieving 97% prediction accuracy. The predictive capabilities of the model contribute to energy savings of up to eightfold, underscoring its practical impact.

By integrating prognostics and self-aware systems into SON, this research demonstrates a pathway to achieve self-optimization and self-healing in 5G and beyond networks, laying the groundwork for sustainable, intelligent network management.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Subjects: T Technology > TK Electrical engineering. Electronics Nuclear engineering
• Colleges/Schools: College of Science and Engineering > School of Engineering
• Supervisor's Name: Imran, Professor Muhammad and Abbasi, Professor Qammer
• Date of Award: 2025
• Depositing User: Theses Team
• Unique ID: glathesis:2025-84868
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 07 Feb 2025 13:08
• Last Modified: 07 Feb 2025 13:08
• Thesis DOI: 10.5525/gla.thesis.84868
• URI: https://theses.gla.ac.uk/id/eprint/84868

Related URLs

• Article DOI
• Conference proceeding
• Conference proceeding
• Book section
• Article DOI