Dynamic Features Adaptation in Networking: Toward Flexible training and Explainable inference

AI systems managing networks need to adapt as conditions change: The paper addresses a critical real-world challenge—as 6G networks evolve with new equipment, vendors, and services, AI models must continuously learn from new data and features rather than remaining static. This flexibility is essential for responsible deployment in critical infrastructure.

Explainability becomes more practical with smart resource allocation: The proposed DAFI method makes AI decision-making more transparent by intelligently deciding when to use computationally expensive explanation techniques versus lighter alternatives. This approach reduces runtime by up to 2 times while maintaining consistency, making explainability feasible for real-time network operations where both speed and interpretability matter.

Iterative learning improves reliability over time: The research demonstrates that Adaptive Random Forests can produce increasingly stable and accurate predictions as they encounter more features and data. This suggests that AI systems managing networks can become more trustworthy through continuous operation, rather than degrading—an important consideration for systems controlling critical infrastructure.

Monitoring for data drift is essential for responsible AI: By detecting when network conditions fundamentally change (distributional drift), the framework signals when AI explanations need refreshing. This prevents the hidden problem of AI systems making decisions based on outdated reasoning, which is crucial for maintaining accountability in network management.