Jessica won a Joint “Best Final-year Project 2025-26” awarded by the School of Computing, University of Kent. She graduated in BSc (Hons) in Computer Science and did a dissertation project on “Enhancing the Explainability of ML-Based NIDS Using XAI and LLM Explanation”, under the supervision of Professor Shujun Li. We want to congratulate Jessica for her remarkable achievement!
This is what Jessica had to say:
Could you tell us about your experiences studying at Kent?
Studying Computer Science at Kent gave me the opportunity to develop my interest in artificial intelligence, particularly deep learning, and gain experience carrying out independent research through my final year project.
I particularly appreciated the support of my final-year project supervisor, Professor Shujun Li. He was open to a project that combined my interest in machine learning algorithms and large language models with a cyber security application, and our regular discussions helped me develop and refine the research. Having that academic freedom, alongside his guidance, made the project a particularly enjoyable experience.
I graduated with a first-class degree and an overall mark of 81.3%. Receiving recognition for my dissertation was a very rewarding way to finish my time at Kent.
Could you tell us about your winning final project?
My project, “Enhancing Explainability of ML-Based NIDS Using XAI and LLM Explanation”, investigated how to make the decisions of machine learning models used in network intrusion detection systems (NIDS) easier to understand for Security Operations Centre workers. I combined machine learning classifiers with SHAP (SHapley Additive exPlanations), an explainable AI methodology that provides insight into how individual data features contribute to a classifier’s prediction. I then used large language models such as GPT-5 and Qwen 3.5 to turn those feature contributions into short explanations in plain English. The aim was to give readers an explanation grounded in the classifier’s decision.
I tested the approach across four network traffic datasets and compared seven language models, assessing the usefulness of their explanations and whether they introduced unsupported information. I also developed an interface to present the predictions and explanations and gathered feedback through a small focus group.
The project made me especially interested in how we evaluate AI explanations. An explanation can sound convincing, but we still need to check whether it accurately reflects the evidence behind the prediction.
What are your plans for the future?
I have started an MSc in Artificial Intelligence and Adaptive Systems at the University of Sussex. I want to build on my undergraduate research and deepen my understanding of deep learning, particularly Large Language Models.