Options
2025
Conference Paper
Title
Proactive 5G Core Network Failure Prediction using Nonlinear Dynamical Analysis and Statistical Outlier Detection
Abstract
The growing complexity of 5G network infrastructure, driven by virtualization and diverse use cases, has made network monitoring increasingly challenging. Traditional Quality of Service (QoS) monitoring tools often overlook the control plane, focusing instead on data plane metrics. This research analyzes the dynamics of a 5G core network to develop a failure prediction mechanism that leverages insights from the control plane. The system's behavior is captured in phase space using Takens' embedding method. However, the analysis rules out the presence of a strange attractor, as the network metrics exhibit a positive correlation. The Lyapunov exponent is also calculated, revealing quasi-periodic divergence of trajectories over time, which indicates a lack of significant chaotic behavior. Based on these findings, a statistical approach for outlier detection is proposed. Metrics from the network's normal state are encapsulated within a hypersphere to compute a centroid, and the Mahalanobis distance measures deviations from this centroid. The density of these outliers is monitored, and if it surpasses a set threshold, the system is flagged as being at high risk of failure. This proactive and computationally efficient approach is highly suitable for real-time failure prediction, ensuring network reliability and service continuity.
Author(s)
Conference