TriSLA: A Preventive and Closed-Loop SLA-Aware Architecture for Multidomain Decision-Making with Explainable Artificial Intelligence in 5G Networks

TriSLA achieves 100% SLA compliance for admitted network slices, completely eliminating post-deployment violations compared to reactive systems (51.2% compliance) and static threshold approaches (80.4% compliance), demonstrating the value of preventive rather than reactive decision-making in 5G networks.

The system uses Explainable AI to make admission decisions transparent and interpretable, employing XAI feature attribution methods (specifically a Random Forest classifier achieving 98.68% accuracy) so that network operators can understand why the system approves or rejects service requests rather than treating AI decisions as a "black box."

Preventive feasibility assessment occurs before infrastructure commitment, with the machine learning model predicting SLA violations at request time with up to 99.51% accuracy, allowing the system to reject risky requests upfront rather than discovering problems after deployment.

The architecture maintains continuous oversight through closed-loop monitoring, combining initial feasibility checks with real-time telemetry collection and runtime assurance to catch and address SLA violations throughout a service's operation, not just at admission.

The system adds minimal computational burden to network operations, requiring only 25.37 ms for semantic parsing and 231.66 ms for explainable AI assessment, suggesting that responsible AI practices (explainability and preventive governance) need not significantly slow down network decision-making.