Abstract
Fibre networks are critical infrastructure but remain vulnerable to service degradations that can escalate into costly outages and customer disruption. This study applies Action Design Research to develop a predictive framework that turns big-data optical telemetry into actionable insights for decision-making. Co-designed with engineers, the pipeline tackles class imbalance and integrates domain-specific features to predict degradations pre-emptively. The tuned model achieved a high degree of accuracy (≈96%) and excellent separability between fault classes (AUC ≥ .98). The results from field trials confirm the business value of this approach, showing a 28% reduction in unplanned outages and a 24% reduction in maintenance dispatches. It gives organisations a powerful tool to prevent outages rather than respond to them, shifting the operational model from a reactive scramble to a proactive assurance strategy. By embedding principles of fairness, privacy, and auditability, the work ensures responsible use of data-driven methods for infrastructure management.
Recommended Citation
Li, Peng; Harrison, Sean; and Sundaram, David, "From Data to Diagnostics: Action Design Research for Fibre
Outage Prediction Using Explainable Machine Learning" (2025). ACIS 2025 Proceedings. 110.
https://aisel.aisnet.org/acis2025/110