3GPP-compliant NWDAF dataset generation for machine learning-based prediction of network metrics
Loading...
Date
2025-10-16
Authors
Φυρογένη, Αριάδνη
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The evolution of 5G networks introduces significant challenges in management and optimization due to their complexity, dynamic resource allocation and diverse service requirements. To address these, network metrics are collected and organized into standardized datasets that enable interoperability and reliable analysis. These datasets form the basis for applying Artificial Intelligence (AI) techniques, which support predictive analytics, anomaly detection and automated decision-making in next-generation networks. In this context, the Network Data Analytics Function (NWDAF), specified by the 3rd Generation Partnership Project (3GPP) in specifications 23.288 and 29.520, serves as the main analytics entity in 5G Core networks. It gathers measurements from various Network Functions (NFs) and provides advanced analytics services for intelligent orchestration. This thesis extends an existing NWDAF implementation by introducing a standardized dataset generation process aligned with 3GPP requirements. The resulting datasets were used in machine learning experiments to predict CPU and memory usage using Long Short-Term Memory (LSTM) models, showcasing the potential of combining standardized NWDAF outputs with AI for improved network performance forecasting. The experimental setup integrated an Open5GS core, an Amarisoft RAN and iPerf-generated UE traffic, with Prometheus collecting metrics and NWDAF producing standardized datasets for model training in PyTorch. The results confirm the feasibility of standardized datasets for enabling AI-driven analytics and demonstrate NWDAF’s practical value in enhancing 5G network management.
Description
Keywords
5G networks, Network data analytics function (NWDAF), Datasets, Machine learning, 3rd generation partnership project (3GPP)