Node-Level KPI Regression over Graph-Structured Mobile Network Data: An Empirical Investigation

• Guilherme F. R. Santos Ericsson Research Brazil / UFSCar
• Amadeu do Nascimento Junior Ericsson Research Brazil

DOI: https://doi.org/10.5753/wiarc.2026.22942

Resumo

6G RAN management demands autonomous solutions translating high-level QoS/QoE objectives into configuration changes with measurable KPI impact. We address KPI prediction from configuration adjustments—a problem largely underexplored outside temporal settings—framed as node-level regression on graph-structured data. Through a principled ablation study, we show that lightweight GNNs outperform flat baselines by capturing topological dependencies, that KPI responses are predominantly local, and that residual connections with dedicated MLP blocks yield consistent gains, while self-supervised pretraining fails to transfer to this setting. Our results provide an empirical foundation for AI/ML-driven autonomous next-generation RAN management.

Referências

Akiba, T., Sano, S., Yanase, T., Ohta, T., and Koyama, M. (2019). Optuna: A nextgeneration hyperparameter optimization framework. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD).

Biyar, E. D. et al. (2025). Autonomous conflict handling in intent-based management. Computer Networks, 271:111561.

Fey, M. and Lenssen, J. E. (2019). Fast graph representation learning with pytorch geometric. In ICLR Workshop on Representation Learning on Graphs and Manifolds.

Gilmer, J., Schoenholz, S. S., Riley, P. F., Vinyals, O., and Dahl, G. E. (2017). Neural message passing for quantum chemistry. In Advances in Neural Information Processing Systems, volume 30, pages 1263–1272. Curran Associates, Inc.

He, K., Zhang, X., Ren, S., and Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

Hexa-X (2023). Deliverable d3.2 - initial architecture enablers. Technical report, Hexa-X.

Hou, Z., Liu, X., Cen, Y., Dong, Y., Yang, H., Wang, C., and Tang, J. (2022). Graphmae: Self-supervised masked graph autoencoders. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining.

Kipf, T. N. and Welling, M. (2017). Semi-supervised classification with graph convolutional networks. In Proceedings of the 5th International Conference on Learning Representations (ICLR).

Lin, J., Lan, T., Zhang, B., Lin, K., Miao, D., He, H., Ye, J., Zhang, C., and Li, Y.-F. (2025). Multi-scenario cellular kpi prediction based on spatiotemporal graph neural network. IEEE Transactions on Automation Science and Engineering, 22:5131–5142.

Rydén, H. et al. (2023). Next generation mobile networks’ enablers: Machine learning-assisted mobility, traffic, and radio channel prediction. IEEE Communications Magazine, 61(10):94–98.

Sha, T., Zhang, Y., Li, Q., Zhang, Z., Zhu, L., Hua, X., Yu, R., Fan, X., Lei, Z., Feng, J., and Zhang, Y. (2024). World model aided parameter adjustment decision and evaluation system for radio access network. In ICC 2024 - IEEE International Conference on Communications, pages 1649–1654.

Silva, A., Temesgene, D. A., Klautau, A., Aben-Athar, R., and Nahum, C. (2025). Leveraging GNNs for intent-driven 5G RAN optimization in autonomous networks. IEEE Access, 13:189096–189110.

Temesgene, D. A., Biyar, E. D., Silva, A., Likhyani, A., and Zahemszky, A. (2024). Graph neural network for building prediction agents in intent-based zero-touch networks. In Proceedings of the IEEE International Conference on Communications (ICC), pages 974–979.