Research on Network Risk Evaluation Method Based on a Differential Manifold

Xiaolin Zhao, Yiman Zhang, Jingfeng Xue, Chun Shan, Zhen LiuIEEE Access 2020 · OpenReview

Abstract

With the rapid development of networks, network security is a serious problem. To evaluate a network accurately, this paper proposes a network risk evaluation method based on a differential manifold (DM) and research on traditional methods. The DM divides the network risk evaluation into network structure risk and network behavior risk evaluations. Network structure risk evaluates the network identity, and network behavior risk evaluates the attack and defense of the network. Network assets and asset vulnerabilities characterize a network, and the analytic hierarchy process (AHP) and the Common Vulnerability Scoring System (CVSS) are combined to evaluate the network identity. Network behavior causes high-dimensional indicator changes, and DMs are used to measure network behavior. To examine the effectiveness and accuracy of DMs, two experiments were performed. The experimental results show that the DM method is valid and accurate for evaluating network risk.

Introduction

Evaluating network risk is a complex challenge due to the dynamic nature of modern cyber environments. Traditional evaluation methods often suffer from excessive subjectivity, isolation of vulnerability scores, or a heavy reliance on large labeled datasets. This research introduces a dual-layered framework that separates risk into Network Structure Risk (static identity) and Network Behavior Risk (dynamic activity), utilizing Differential Manifolds (DM) to provide a mathematically rigorous and objective evaluation.

Methodology

Network Structure Risk Evaluation

The static risk of a network is determined by its identity, which includes its assets and their inherent vulnerabilities. To quantify this, the researchers combine the Analytic Hierarchy Process (AHP) with the Common Vulnerability Scoring System (CVSS).

AHP is used to build a hierarchical model of network assets, ensuring that critical infrastructure (like core servers) is weighted more heavily than peripheral devices. This hierarchical approach addresses the subjectivity of expert weighting by providing a structured comparison framework.

The Analytic Hierarchy Process (AHP) structure used for asset weighting.

Simultaneously, CVSS scores are integrated to characterize the vulnerabilities of these assets. By grouping metrics into Base, Temporal, and Environmental categories, the system can more accurately reflect the potential impact of a vulnerability within a specific network context.

The Common Vulnerability Scoring System (CVSS) metric groups integrated into the risk evaluation.

Network Behavior Risk Evaluation

The core innovation of this paper is the use of Differential Manifolds (DM) to model dynamic network behavior. Network indicators—such as CPU usage, packet loss, and flow rates—are treated as coordinates in a high-dimensional space. The "state" of the network is represented as a point on a manifold.

When an attack occurs, the network state moves, creating a trajectory. By applying Riemannian geometry, the researchers calculate the "geodesic distance" between the normal operating state and the current state. This distance represents the "utility" ($U$) of the attack. If $U$ exceeds a predefined baseline ($\Delta$), the network is flagged as being under risk.

Experimental Validation

The model was validated using two distinct experimental setups:

  1. Custom Simulation: A controlled environment with eight hosts was used to test specific Denial of Service (DoS) attacks, such as UDPFlood and SYNFlood.
  2. CICIDS2017 Dataset: A standard benchmark containing a mix of benign traffic and various attacks (Heartbleed, DDoS, Infiltration, etc.) was used to test the model's real-world applicability.

Results and Discussion

The experimental results confirm that the DM-based utility score is a highly sensitive indicator of network risk. In simulation tests, UDPFlood attacks produced utility values nearly 30 times the baseline. In the CICIDS2017 dataset, most attacks resulted in utility values more than 6 times the baseline.

The geometric approach is particularly effective at detecting traffic-intensive attacks that cause significant shifts in network resource indicators. However, "low-and-slow" attacks, such as PortScans, proved more challenging to distinguish from normal background noise, as they create smaller "warps" in the manifold geometry.

Conclusion

By treating network security as a geometric problem on a differential manifold, this research provides a novel way to quantify dynamic risk without relying on subjective expert opinions or massive labeled datasets. Future work will focus on optimizing the Riemannian metrics used and integrating network topology more deeply into the manifold calculations to improve detection of stealthy attacks.

Summary updated

Abstract

With the rapid development of networks, network security is a serious problem. To evaluate a network accurately, this paper proposes a network risk evaluation method based on a differential manifold (DM) and research on traditional methods. The DM divides the network risk evaluation into network structure risk and network behavior risk evaluations. Network structure risk evaluates the network identity, and network behavior risk evaluates the attack and defense of the network. Network assets and asset vulnerabilities characterize a network, and the analytic hierarchy process (AHP) and the Common Vulnerability Scoring System (CVSS) are combined to evaluate the network identity. Network behavior causes high-dimensional indicator changes, and DMs are used to measure network behavior. To examine the effectiveness and accuracy of DMs, two experiments were performed. The experimental results show that the DM method is valid and accurate for evaluating network risk.
BibTeX
@inproceedings{zhao2020research,
  author = {Xiaolin Zhao and Yiman Zhang and Jingfeng Xue and Chun Shan and Zhen Liu},
  title = {{Research on Network Risk Evaluation Method Based on a Differential Manifold}},
  booktitle = {IEEE Access},
  year = {2020},
  url = {https://openreview.net/forum?id=By12ikr5Qb},
}
Figure