Device types affect fairness and accuracy in federated location prediction
Device-Induced Mobility Heterogeneity and Its Impact on Federated Next-Location Prediction
Computers and SocietyNetworking and Internet Architecture
Summary
Predicting where people will go next is useful but challenging when data is kept private on each person's device. The authors studied how different smartphone models show different patterns of movement, which can bias predictions. They found that grouping devices by type and balancing these groups when training models helps make predictions fairer without losing accuracy. Another method also improves fairness but not as much, and works when device info is missing.
What this means in practice
- •For mobile network operators: Improve next-location prediction models by balancing device types in federated training, leading to fairer and accurate mobility predictions.
- •For urban planners: Use improved federated models to better understand mobility patterns without accessing raw user data, assisting in privacy-conscious urban planning.
Authors
André Felipe Zanella, Paula Delgado de Santos, José Suárez-Varela, David Solans Noguero, Andra Lutu, Jesus Omaña-Iglesias
Abstract
Predicting human mobility is central to network management, urban planning, and location-based services, yet privacy concerns are increasingly pushing such tasks toward federated learning. A key challenge in this setting is the strong heterogeneity of mobility data across users, which leads to non-IID data distributions and can significantly impact model performance and fairness. In this paper, we present a large-scale measurement study of device-level mobility heterogeneity using data from a major MNO in a dense European capital. Our dataset comprises mobility traces derived from network events, covering over 50,000 UEs across 178 smartphone models. We group devices into tiers based on their market segment and show that mobility characteristics, such as spatial range and regularity, vary systematically across these tiers, revealing a previously underexplored source of bias in mobility data. We investigate the implications of this heterogeneity on the use case of federated next-location prediction. In particular, we find that balancing device tiers across silos through tier-aware client selection yields the best results, improving fairness by up to 34.49% without degrading accuracy. In contrast, a FedProx-only approach achieves a 21.46% fairness improvement, offering a reasonable alternative when device information is unavailable, though with lower overall gains.