BROWSE
Detail
Conference Paper
Design of a Post-hoc XAI Function Architecture for Intelligent Threat Detection Systems in P5G Networks
Cited - time in 
Share 
Share
- Authors
- Mikyong Han, Noh-Sam Park, Yongyoon Shin, Hyunjin Kim, Jonghoon Lee
- Issue Date
- 2026-10
- Citation
- International Conference on Information and Communication Technology Convergence (ICTC) 2026, pp.1-4
- Publisher
- IEEE
- Language
- English
- Type
- Conference Paper
- Abstract
- Private 5G (P5G) network security increasingly relies on self-supervised learning (SSL)-based threat detection models to address the scarcity of labeled attack data. However, the black-box nature of these models presents significant challenges in developing AI models optimized for network intrusion detection systems (NIDS), particularly in SSL frameworks that support heterogeneous encoders, and in enabling downstream response systems to logically interpret inference results. These limitations have increased the need for explainable artificial intelligence (XAI) in SSL-based network security. To address these challenges, this paper reviews recent advances in XAI techniques for network intrusion detection and proposes an XAI functional architecture applicable to SSL-based detection models supporting heterogeneous encoders. The proposed architecture is designed to support the development and optimization of AI models for NIDS while providing interpretable explanations that enable downstream response systems to understand and utilize inference results. It integrates a model-agnostic post-hoc analyzer with a threat explanation generator to transform XAI outputs into human-readable natural-language explanations. In addition, the architecture incorporates a continuous feedback and retraining loop to progressively improve explanation quality over time. This paper presents the functional design of the proposed XAI architecture and discusses its applicability to real-time intrusion detection in P5G network environments.
- Keyword
- Private 5G (P5G), Intrusion Detection System (IDS), Self-Supervised Learning (SSL), Post-hoc Explanation, Explainable AI (XAI), Feature Attribution
- KSP Keywords
- Black box, Continuous feedback, Functional Architecture, Functional design, Human-readable, Intrusion Detection Systems(IDS), Network Security, Over time, Real-time intrusion detection, Threat Detection, artificial intelligence