AutoML in the Face of Adversity: Strategies for Mobility Predictions in NWDAF

This research, authored by Syafiq al Atiiq, Christian Gehrmann, and Yachao Yuan from Lund University, and Jakob Sternby from Ericsson Research, is part of the ELASTIC Project and was presented at the 9th International Conference on Fog and Mobile Edge Computing (FMEC 2024).

About the Research
The study investigates the Network Data Analytics Function (NWDAF), a key component of 5G networks as defined by 3GPP standards. NWDAF uses machine learning models to optimize network performance, requiring periodic retraining to maintain accuracy. Adversarial user equipment (UE), however, can introduce corrupted data during retraining, compromising model reliability.

Two retraining strategies were evaluated using Automated Machine Learning (AutoML):

1. Reselecting the model during each retraining.
2. Retaining a well-optimized initial model.

The findings recommend prioritizing model optimization during the initial training phase to maintain prediction accuracy, even when retraining processes are impacted by adversarial data.

Access and Further Information
The full publication and supporting materials are available on Zenodo.

This research offers insights into improving the robustness of mobility predictions in 5G networks.