motc/traffic-flow-predictor
Traffic Flow Predictor
MAE (veh/15min)
21.4
MAPE
9.2%
RMSE (veh/15min)
34.8
R²
0.91
About this model
Graph neural network that predicts vehicle flow on 214 monitored corridors across Greater Doha at 15-minute horizons up to two hours ahead. Ingests live inductive-loop and radar sensor feeds and accounts for weekday, Ramadan, and school-calendar effects.
Intended use
Signal-timing optimisation, corridor management dashboards, and pre-event traffic planning by MOTC and municipal operations teams.
Training data lineage
Trained on three years of traffic-sensor-feeds observations fused with vehicle-movement-aggregates for corridor-level calibration.
Limitations & bias notes
Accuracy degrades during unplanned closures and severe dust storms when sensor dropout exceeds 20%. Corridors instrumented after January 2026 rely on transferred weights and run roughly 30% higher error until retraining.
Evaluation metrics
| MAE (veh/15min) | 21.4 |
| MAPE | 9.2% |
| RMSE (veh/15min) | 34.8 |
| R² | 0.91 |
Try it
Live sandboxhorizon=12&profile=weekday_peak
fixtureFeedback
Owning entity
- Updated
- 2026-07-08
- Latest version
- 2.1.0
- License
- Open
- Access
- Open