Intelligent QoS Prediction in Software-Defined Networking Via Lightweight Graph-Temporal Machine Learning

Author (s)

Chernor Illah Jalloh, Julliet Josiah, Christiana Lebbie, Tom Penyikie, & Ishmael Sheriff

Abstract

Timely Quality of Service (QoS) prediction can enable software-defined networks to identify emerging performance problems before users experience excessive delay, congestion, packet loss, or unstable service. This study presents LiGHT-QoS, a lightweight graph-temporal forecasting framework that combines port-level network topology and telemetry with recent end-to-end QoS measurements. The experimental dataset was generated from 120 independent Mininet and Open vSwitch runs involving linear, tree, and mesh topologies, eight operating conditions, and five repetitions for each topology and condition. Following timing validation and segment-preserving filtering, 33,152 of 36,000 one-second intervals were retained, representing a 92.09% acceptance rate and 375,868 node-level observations. Model performance was evaluated at 1, 3, and 5-second forecasting horizons using five grouped folds and five random seeds, while ensuring that data from the same run did not appear in both training and testing sets. Compared with persistence forecasting, the proposed HGTE model reduced scale-balanced macro means absolute error by 13.30%, 13.40%, and 13.00% at the three respective horizons. HGTE achieved the best overall ranking at 1 and 3 seconds and was only 0.28% behind a QoS-only GRU at 5 seconds. Jitter prediction showed the most consistent improvement, with mean absolute error reductions between 21.07% and 22.11%. The model contains only 12,452 trainable parameters and achieved an 82.63% reduction in median CPU inference latency after export to ONNX FP32. These findings demonstrate that LiGHT-QoS can support efficient controller-assisted network monitoring, although real-network validation and calibrated fallback mechanisms remain necessary before large-scale operational deployment.

Key words: Software-defined networking, Quality of service, Graph neural network, Temporal forecasting, Topology generalization, Machine Learning, Edge inference.

Download PDF

Cite This Article:

Jalloh, C. I., Josiah, J., Lebbie, C., Penyikie, T., & Sheriff, I. (2026). Intelligent QoS Prediction in Software-Defined Networking Via Lightweight Graph-Temporal Machine Learning, Journal of Scientific Reports, 15(1), 283-303. https://doi.org/10.58970/JSR.1251

About Author (s)

Chernor Illah Jalloh (Corresponding Author), School of Electronics Information and Communication, Huazhong University of Science and Technology, Wuhan, China. ORCID: https://orcid.org/0009-0002-4803-1474

Julliet Josiah, School of Electronics Information and Communication, Huazhong University of Science and Technology, Wuhan, China. ORCID: https://orcid.org/0009-0007-8079-2026

Christiana Lebbie, School of Electronics Information and Communication, Huazhong University of Science and Technology, Wuhan, China. ORCID: https://orcid.org/0009-0001-7110-3651

Tom Penyikie, School of Electronics Information and Communication, Huazhong University of Science and Technology, Wuhan, China. ORCID: https://orcid.org/0009-0006-6045-0427

Ishmael Sheriff, School of Electronics Information and Communication, Huazhong University of Science and Technology, Wuhan, China; School of Technology, Western Governors University (WGU), 4001 South 700 East, Suite 700 Salt Lake City, Utah 84107 United States.  ORCID: https://orcid.org/0009-0009-4653-5485

Download PDF

DOI: https://doi.org/10.58970/JSR.1251

Summarize this article with: