Device differences impact how next-location prediction works
Mobility data varies across 178 smartphone models and affects federated learning performance.
Mobility characteristics like spatial range and regularity vary across device tiers. This device-induced mobility heterogeneity impacts federated next-location prediction.
A study used mobility traces from over 50,000 UEs across 178 smartphone models. The data came from a major MNO in a dense European capital.
Federated learning is a way to perform tasks while addressing privacy concerns. Next-location prediction is a task used for network management and urban planning.
Tier-aware client selection improves fairness by up to 34.49% without degrading accuracy. This method works by balancing device tiers across silos.
A FedProx-only approach achieves a 21.46% fairness improvement. This is a reasonable alternative when device information is unavailable.
Key facts
- Mobility data shows strong heterogeneity across different device tiers.
- Tier-aware client selection improves fairness by up to 34.49% without degrading accuracy.
- The study included data from over 50,000 UEs and 178 smartphone models.