A Method for Calculating Network System Security Risk Based on a Lie Group

Xiaolin Zhao, Quanbao Chen, Jingfeng Xue, Yiman Zhang, Jingjing ZhaoIEEE Access 2019 · OpenReview

Abstract

Traditionally, network risk assessment uses a statistical computation method. This paper proposes Lie group kinematics to describe the feature space of the attack behavior. A matrix composed of indicators and topologies in a network system is mapped to a Lie group. The attack behavior path and the powers of the attack and defense are defined. The risk value of the attack and defense is calculated from the change in the indicators. Then, we evaluate the network security risk status using calculus. To examine the validity of the network risk assessment based on a Lie group, we conduct one experiment and utilize the CIC2017 dataset to show the applicability and efficiency of the proposed method. The experimental results show the effectiveness of the calculation method based on a Lie group, and the risk value of the attack and defense is valid compared to those of other machine learning algorithms. The calculation method based on a Lie group can quantitatively analyze network security risks.

Introduction

Traditional network risk assessment often relies on static statistical models. While useful, these methods struggle with two major issues: state explosion and dynamic blindness. In complex networks with thousands of nodes, the number of possible security states grows exponentially, making exhaustive modeling impossible. Furthermore, attacks are not single events; they are processes that evolve over time. Static models often fail to capture the "momentum" or direction of an ongoing attack.

This paper aims to bridge this gap by providing a mathematical framework that describes the kinematics of an attack—how the "position" of a network’s security state shifts in a high-dimensional space as an attacker moves through the system. By moving beyond traditional statistical risk assessment, this research models network security through the lens of differential geometry.

Methodology

The core innovation of this research is the application of Lie groups—mathematical structures representing continuous symmetries and transformations—to network topology and performance indicators. The research process follows a structured path from theoretical definition to experimental verification.

Flowchart of the proposed research process, from theoretical Lie group definitions to experimental verification.

1. Matrix Mapping and Differential Manifolds

The network state is represented as a matrix composed of indicators (such as CPU usage, memory, and traffic) and the physical topology. These matrices form a General Linear Group, a type of Lie group. By treating the network state as a point on a smooth manifold, an attack becomes a continuous "path" or trajectory. The "distance" moved along this path (geodesic distance) represents the quantitative change in risk.

2. Utility Functions and Calculus-Based Risk

A utility function is defined as $\mu(x) = E_{in} - E_{out}$, where $E_{in}$ represents the "attack energy" (damage potential and probability) and $E_{out}$ represents the "defensive power." Instead of a static snapshot, the model uses calculus to integrate the change in indicators along the attack path, providing a real-time risk value. This mapping makes the data "dimensionless," allowing the model to handle disparate indicators like bits per second and CPU load within the same geometric framework.

Experimental Design

The model was validated using two primary sources: a custom DDoS experiment and the benchmark CICIDS2017 dataset. The custom experiment utilized a controlled environment to simulate the continuous nature of a Distributed Denial of Service attack.

The network topology used for the custom DDoS experiment, featuring a hacker, firewall, and various server targets.

As shown in the topology, the attack path moves from the external hacker through a firewall and routing infrastructure to reach critical assets like Web and Redis servers. This setup allowed the researchers to track the "motion" of the network state as indicators fluctuated during the attack.

Results and Discussion

The Lie group method achieved a Precision, Recall, and F1-score of 0.83. While this was lower than top-tier machine learning models like Random Forest (0.97), it represents a successful "medium-level" performance for a novel mathematical approach.

The primary advantage of this method is computational efficiency. Because it relies on $O(n)$ linear algebra and calculus rather than the $O(n^2)$ comparisons required by algorithms like K-Nearest Neighbors (KNN), it is significantly more scalable. In testing, the execution time was approximately 20 seconds, vastly outperforming KNN (1908 seconds) and Adaboost (1126 seconds) on identical hardware.

However, limitations remain. The model excels at tracking continuous changes (like DDoS) but is less effective for "discrete" events like port scans, where the change in the network's geometric position is less pronounced. Additionally, the current utility functions are simplified; real-world application would require more complex, non-linear functions to represent attack and defense energy accurately.

Conclusion

Mapping network security to Lie groups provides a robust mathematical foundation for quantitative risk analysis. By treating attacks as continuous motion, the framework enables efficient, real-time assessment that scales effectively with network size. Future work will focus on optimizing node weights and refining utility functions to improve detection accuracy for discrete attack patterns.

Summary updated

Abstract

Traditionally, network risk assessment uses a statistical computation method. This paper proposes Lie group kinematics to describe the feature space of the attack behavior. A matrix composed of indicators and topologies in a network system is mapped to a Lie group. The attack behavior path and the powers of the attack and defense are defined. The risk value of the attack and defense is calculated from the change in the indicators. Then, we evaluate the network security risk status using calculus. To examine the validity of the network risk assessment based on a Lie group, we conduct one experiment and utilize the CIC2017 dataset to show the applicability and efficiency of the proposed method. The experimental results show the effectiveness of the calculation method based on a Lie group, and the risk value of the attack and defense is valid compared to those of other machine learning algorithms. The calculation method based on a Lie group can quantitatively analyze network security risks..
BibTeX
@inproceedings{zhao2019method,
  author = {Xiaolin Zhao and Quanbao Chen and Jingfeng Xue and Yiman Zhang and Jingjing Zhao},
  title = {{A Method for Calculating Network System Security Risk Based on a Lie Group}},
  booktitle = {IEEE Access},
  year = {2019},
  url = {https://openreview.net/forum?id=Q6P4RfET4A},
}
Figure