Hussein Fawaz, Farhad Arpanaei, Davide Andreoletti, Ihab Sbeity, José A. Hernández, David Larrabeiti, Omran Ayoub
Lebanese University, SUPSI
International Conference on Optical Network Design and Modeling (ONDM), 2024
The problem
Quality-of-transmission (QoT) estimation models for lightpaths in optical networks often rely on large feature sets. This increases computational cost and limits interpretability — both problematic in operational optical networks.
What we propose
We use SHAP values to quantify feature importance and systematically remove redundant inputs, building simpler and more accurate ML models for lightpath QoT estimation. This reduces model complexity, speeds up inference, and improves interpretability.
Results and takeaways
Our method achieves higher predictive accuracy with smaller feature sets and better generalization. It shows how explainable AI can be leveraged not only for interpretation, but also for practical system optimization. Next, we aim to apply this framework to other networking problems and integrate reliability-aware feature selection.
