Proactive management of 5G networks requires accurate forecasting of key performance measurements (KPMs), but existing machine learning approaches face significant limitations in scalability and explainability, restricting their effectiveness in real-world deployments. A new framework called TelcoAgent, detailed in an arXiv paper (ID: 2606.19821), aims to overcome these hurdles by combining foundation models with 3GPP-grounded reasoning.
The Scalability and Explainability Challenge
Current ML approaches for KPM forecasting often require site-specific training and lack transparent reasoning, making them difficult to deploy across diverse network cells and hard for operators to trust. TelcoAgent is designed as a foundation model-based framework that enables accurate, scalable, and explainable forecasting of multiple KPMs across diverse network cells without the need for site-specific training, according to the authors.
TelcoAgent's Three-Component Architecture
The TelcoAgent framework comprises three key components, each addressing a critical aspect of the forecasting workflow:
| Component | Description |
|---|---|
| Automated three-agent pipeline | Constructs a 3rd Generation Partnership Project (3GPP) knowledge graph directly from specification documents, grounding the system in telecom standards. |
| Scalable time-series foundation model (TSFM)-based prediction pipeline | Delivers accurate zero-shot forecasting without requiring per-cell model retraining. |
| Reasoning and explanation pipeline | Provides actionable, domain-grounded diagnostics and actionable instructions to address network degradations. |
The use of a 3GPP knowledge graph ensures that predictions are rooted in official telecom specifications, enhancing both accuracy and trust.
Evaluation on Real-World 5G Data
The researchers evaluated TelcoAgent using a 3-month, real-world, city-scale 5G KPM dataset from a U.S.-based network operator. The framework demonstrated high forecasting accuracy for all 7 considered KPMs per cell across 200 cells, while simultaneously delivering explainable insights and actionable instructions to diagnose and address network degradations. This zero-shot capability eliminates the need for site-specific training, dramatically improving scalability.
Implications for Network Operations
By providing both accurate forecasts and explainable diagnostics, TelcoAgent enables network operators to proactively manage degradations without manual analysis. Its grounding in 3GPP standards ensures that recommendations are aligned with industry best practices. For CTOs and telecom leaders, this represents a shift from black-box ML models to transparent, scalable AI systems that can be deployed across entire networks. The framework's ability to operate without site-specific training means faster rollouts and lower operational costs, potentially reshaping how 5G and next-generation networks are monitored and optimized.