Abstract—The increasing dynamism of modern communication infrastructures, driven by technologies such as 5G, edge computing, and Software-Defined Networking (SDN) necessitates predictive intelligence capable of modeling temporal variations in topology and traffic. We present RouteNet-TGNN, a novel temporal graph neural network that extends the foundational RouteNet framework by incorporating Gated Recurrent Units (GRUs) within its path-based message-passing architecture. This integration enables the model to capture both spatial dependencies across network paths and temporal dynamics over sequential states, facilitating accurate multi-step forecasting of critical Quality of Service (QoS) indicators including delay, jitter, and packet loss. Experimental evaluations across diverse synthetic network scenarios demonstrate that RouteNet-TGNN achieves substantial gains over static GNN baselines, reducing delay prediction error by 27.5%, jitter estimation error by 21.3%, and packet loss prediction error by 18.9%. Designed for short-term predictive horizons (t+1 to t+3), the proposed model enhances operational visibility and enables proactive decision-making in SDN controllers and digital twin environments. These findings underscore the potential of spatio-temporal GNN architectures to advance autonomous, adaptive, and resilient network management.


Keywords—Spatio-temporal graph neural networks, RouteNet-TGNN, temporal forecasting, Quality of Service prediction, Software-Defined Networking (SDN) analytics, network intelligence, Gated Recurrent Unit (GRU) integration, proactive network management

Cite: Sultan Ahmad, Gadu Srinivasa Rao, Hikmat A. M. Abdeljaber, Eali Stephen Neal Joshua, Faroug A. Abdalla, and Himaja Gadi, “RouteNet-TGNN-A Temporal Graph Neural Network for Delay and Loss Forecasting in Dynamic Communication Networks," Journal of Communications, vol. 21, no. 2, pp. 211-225, 2026.

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).