Calibrated Network Security Situation Prediction Based on Neural Granger Causal Representation and Bayesian Graph Neural Networks
Concurrency and Computation: Practice and Experience · 2026-05 · 38 (10)
DOI: 10.1002/cpe.70760 ↗Kurz gefasst
Fachartikel in der Zeitschrift Concurrency and Computation: Practice and Experience, online erschienen am 18.05.2026. Verfasst von Peng Zhang und 3 weiteren Autor:innen. Die Arbeit nennt 26 Literaturangaben.
Abstract (Originalsprache)
As network attack methods become increasingly complex and covert, traditional passive security defense mechanisms struggle to address the challenges posed by the dynamic evolution of network security situations. This paper proposes a novel network security situation prediction framework based on neural Granger causal representation and Bayesian Graph Neural Networks (Causal‐BGNN) to achieve high‐precision, well‐calibrated situation awareness and threat warning. The method first utilizes a neural Granger causal discovery module to automatically learn sparse causal structures among security indicators from multivariate time series. These structures are then embedded into a Bayesian Graph Convolutional Network, which provides probabilistic uncertainty estimates alongside its predictions. To further enhance operational utility, we design a causal counterfactual interpreter to improve model interpretability, offering actionable guidance for security decision‐making. Comprehensive experiments on three public datasets (CIC‐IDS2017, UNSW‐NB15, and CTU‐13) demonstrate that causal‐BGNN significantly outperforms traditional machine learning, deterministic deep learning, and standard graph neural network baselines. Regarding calibration, the framework achieves highly reliable confidence estimation following temperature scaling. Furthermore, it demonstrates excellent out‐of‐distribution generalization, providing robust detection and early warning capabilities against unknown attacks. This study not only improves the accuracy, calibration, and interpretability of security situation prediction but also provides robust methodological support for building next‐generation trustworthy and adaptive intelligent defense systems.
Zitieren
APA
Zhang, P., Zhao, G., Wu, X., & Wei, L. (2026). Calibrated Network Security Situation Prediction Based on Neural Granger Causal Representation and Bayesian Graph Neural Networks. Concurrency and Computation: Practice and Experience, 38(10). https://doi.org/10.1002/cpe.70760
Vancouver
Zhang P, Zhao G, Wu X, Wei L. Calibrated Network Security Situation Prediction Based on Neural Granger Causal Representation and Bayesian Graph Neural Networks. Concurrency and Computation: Practice and Experience. 2026;38(10). doi:10.1002/cpe.70760
BibTeX
@article{Zhang2026,
author = {Zhang, Peng and Zhao, Guosheng and Wu, Xiaosheng and Wei, Luqi},
title = {Calibrated Network Security Situation Prediction Based on Neural Granger Causal Representation and Bayesian Graph Neural Networks},
journal = {Concurrency and Computation: Practice and Experience},
year = {2026},
volume = {38},
number = {10},
doi = {10.1002/cpe.70760}
}
EUVIMED