Master of Science
This thesis identifies the research gaps in the field of network intrusion detection and network QoS prediction, and proposes novel solutions to address these challenges. Our first topic presents a novel network intrusion detection system using a stacking ensemble technique using UNSW-15 and CICIDS-2017 datasets. In contrast to earlier research, our proposed novel network intrusion detection techniques not only determine if the network traffic is benign or normal, but also reveal the type of assault in the flow. Our proposed stacking ensemble model provides a more effective detection capability than the existing works. Our proposed stacking ensemble technique can detect 90.4% and 98.7% cyberattacks with an f1-score of 90.0% and 98.5%, respectively. Our second topic proposes a novel QoS prediction model tested in a live 5G network environment. Compared to the existing work in this domain, our study is the first approach to conduct a large-scale field test in a 5G network to measure and forecast the network QoS metrics. More than 50 days of continuous data have been collected, cleaned, and used for training the deep sequence models to predict the 5G network QoS metrics such as throughput, latency, jitter, and packet loss. Our experiments demonstrate the effectiveness of predicting the QoS metrics using LSTM and LSTM Encoder-Decoder models, providing lower prediction errors of 14.57% and 13.75%, respectively.
Summary for Lay Audience
The recent advancement of communication technology, such as 5G, has opened up opportunities for next-generation digital services and applications in various sectors such as autonomous and connected vehicles, autonomous drones, smart grid, e-health, and many other smart-city applications. To ensure the uninterrupted operations of such networks, adequate cyber security measures against network intrusions and forecasting network health status are important to the Internet Service Provider (ISP) / network operators. In this thesis, a novel AI-based cyberattack detection methodology and a network quality of service (network health status) prediction framework have been developed and validated. These contributions are presented under two related topics: the first article introduces a network intrusion detection system that uses different deep learning models to detect cyberattacks. The second topic comprises two parts: the first part proposes a network QoS analyzer tool to collect 5G network QoS data, including throughput, latency, jitter, and packet loss, from a live 5G network; and the second part presents a novel network QoS prediction strategy by utilizing deep sequence models. Our proposed models are expected to assist the network planners and operations team in ensuring their committed service level agreement (SLA) to their customers.
Sayem, Ibrahim Mohammed, "Exploring Artificial Intelligence (AI) Techniques for Forecasting Network Traffic: Network QoS and Security Perspectives" (2022). Electronic Thesis and Dissertation Repository. 8861.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License