计算机科学
人工智能
入侵检测系统
图形
人工神经网络
模式识别(心理学)
数据挖掘
特征(语言学)
机器学习
入侵
作者
Bommy M,M. Keerthana Evanjilin,Mutra Jahnavi,C. Keerthi Padma Priya,P. G. Lokeshwari
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-06-01
卷期号:: 1-5
标识
DOI:10.1109/icdca69396.2026.11620243
摘要
An Intrusion Detection System (IDS) operates as a fundamental security solution which monitors and examines network traffic flow. The system's goal is to detect malicious activities that pose threats to its operational integrity. Graph-based intrusion detection systems have recently attracted considerable attention, with network entities such as hosts and communication flows depicted as nodes and edges, allowing Graph Neural Networks (GNNs) to proficiently capture intricate structural and relational patterns. Although current graph-based IDS models successfully detect threats, their decision-making processes remain unclear, which prevents security analysts from trusting the model's predictions. The current limitation requires the development of an intrusion detection system that can provide clear explanations of its detection results. The present research proposes an Explainable Graph-Based Intrusion Detection System (X-GraphIDS) which enhances the original self-supervised GraphIDS model through the introduction of a GNNExplainer-based interpretability module. The proposed technique maintains high detection accuracy while providing node-level explanations for detected anomalies. The system lets you keep track of how decisions are made based on the specific factors that affect them. We have the NF-UNSW-NB15-v3 dataset to see how well the proposed X-GraphIDS model works by looking at standard measures like Accuracy, Precision, Recall, and F1-score. Posthoc explainability methods assess detection performance while they examine how different features impact incorrect predictions. The experimental results show that explainability implementation increases detection transparency while maintaining detection accuracy, which provides a clear and precise method for network intrusion detection in modern network environments
科研通智能强力驱动
Strongly Powered by AbleSci AI