Baku Metro FedAvg GRU
Summary
Federated GRU baseline for station-level next-day passenger entry forecasting under the one-station-one-client setting.
Task
Station-level next-day passenger entry forecasting.
Training Data
Processed dataset reference: olaflaitinen/baku-metro-station-flow
Architecture
Compact GRU regressor trained with Flower FedAvg aggregation under the one-station-one-client setting.
Training Procedure
Clients train local GRU updates on station-specific sequences. The server aggregates model parameters with FedAvg across deterministic training rounds.
Evaluation
Metrics are exported from the local artifact bundle and include MAE and RMSE on held-out dates.
Intended Use
- reproducible federated learning baselines
- transportation demand forecasting experiments
Out-of-Scope Use
- live passenger operations
- identity analytics
- route inference
Ethical and Operational Considerations
This model operates on aggregated station-level entry counts and should not be interpreted as a claim about individual passengers or operational causality.
Reproducibility
See the repository configuration files, requirements lock, and artifact manifest for exact experimental settings.
License
The repository code, documentation source, configuration, and release metadata are licensed under Apache-2.0. The trained model artifact in this repository is released under the same project license unless external platform terms impose additional distribution mechanics.
Citation
Please cite the repository software record in CITATION.cff.
