Explainable AI Intrusion Detection System: Improving Transparency and Trust in Cybersecurity
Nature Journal of Emerging Sciences Technologies and Innovations · 2025
DOI: 10.65752/e5fa2b27 ↗At a glance
Journal article in Nature Journal of Emerging Sciences Technologies and Innovations, 2025. Written by James Onyilo and 1 co-authors. Cited 1 times so far (Crossref). The full text is freely available (open access, CC BY 4.0).
Abstract (original language)
An Explainable Artificial Intelligence (XAI) solution is presented for Intrusion Detection Systems (IDS) that would enhance the transparency, trust, and efficiency of cybersecurity operations. With the increasing complexity of infrastructures, explainability in IDS has been an urgent need since organizations require justifiable decisions made by artificial intelligence. This is particularly important in controlled industries like finance, medical, and military, where it is as important to know the logic behind the alerts as much as detection accuracy. Nonetheless, the traditional deep learning models, despite their effectiveness in network threat detection, are “black boxes,” that is, they provide minimal insights into their decision-making. This non-transparency thus leads to a lack of user confidence and prevents adoption in Security Operations Centers (SOCs). Thus, to address this challenge, a hybrid deep learning model was created with the help of post-hoc explanatory tools like SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-Agnostic Explanations). In addition, SHAP gives a universal insight into the behavior of the model, whereas LIME describes individual predictions, which allows analysts to visualize feature contributions and explain anomalies. Moreover, the model was tested on three benchmark datasets, namely NSL-KDD, UNSW-NB15, and CICIDS2017, and each of them reflects various network threats, including DoS, brute-force, and infiltration attacks. After removing bias through preprocessing, the XAI-IDS reached a recall rate of 98.4% on the CICIDS2017 dataset, which is better than the results from the baseline models. Besides that, SHAP and LIME visualizations also singled out such key features as the Source Bytes, the Destination Port, and the Flow Duration as the most influential. As a result, more trust and less cognitive load were observed in user-centered testing, which corresponds to such ethical artificial intelligence principles as the EU AI Act (2025) and the NIST AI Risk Management Framework (2023).
Cite
APA
Onyilo, J., & Uzuegbu, V. C. (2025). Explainable AI Intrusion Detection System: Improving Transparency and Trust in Cybersecurity. Nature Journal of Emerging Sciences Technologies and Innovations. https://doi.org/10.65752/e5fa2b27
Vancouver
Onyilo J, Uzuegbu VC. Explainable AI Intrusion Detection System: Improving Transparency and Trust in Cybersecurity. Nature Journal of Emerging Sciences Technologies and Innovations. 2025. doi:10.65752/e5fa2b27
BibTeX
@article{Onyilo2025,
author = {Onyilo, James and Uzuegbu, Victoria Chibuzo},
title = {Explainable AI Intrusion Detection System: Improving Transparency and Trust in Cybersecurity},
journal = {Nature Journal of Emerging Sciences Technologies and Innovations},
year = {2025},
doi = {10.65752/e5fa2b27}
}
EUVIMED