V3I10P2

A Topological Data Analysis Approach for Robust Anomaly Detection in Dynamic Computer Networks

G. Yuvaroopa Lakshmi¹, K. Saraswati^2*

Abstract

Modern computer networks generate massive volumes of high-dimensional, noisy, and rapidly changing data, making traditional statistical anomaly detection methods highly susceptible to false positives. To address this challenge, this paper introduces a novel framework utilizing Topological Data Analysis (TDA) to identify structural anomalies in dynamic network traffic. By modeling network interactions as time-evolving graphs, we leverage persistent homology to extract robust topological features—such as connected components (H₀) and multi-dimensional loops (H₁)—that persist across multiple spatial and temporal scales. These topological signatures are then mapped into stable vector spaces using persistence landscapes and integrated into an optimized unsupervised machine learning pipeline. We evaluate our approach on standard network benchmark datasets, demonstrating that our TDA-driven framework outperforms conventional graph neural networks (GNNs) and statistical baselines in detecting low-footprint, distributed cyber-attacks. Crucially, our method maintains structural invariance against high levels of ambient background noise, offering a mathematically rigorous and highly resilient solution for next-generation network security.

Keywords:

Topological Data Analysis, Persistent Homology, Network Anomaly Detection, Cyber-Security, Unsupervised Learning