Plain-language summary
- What it solves
- Compare when to leave and which route alternative is more robust when traffic, weather and uncertainty make average travel time insufficient.
- Who it may help
- People studying forecasting, routing and mobility/logistics decision support in educational scenarios.
- What it produces
- Departure window, route alternative, on-time probability, P10/P50/P90, stress test and calibration metrics.
Show how traffic-aware routing, calibration and uncertainty can support a risk-aware departure choice in commuting scenarios while keeping the project explicitly educational and non-operational.
3 · What it produces
Selected route and departure window, P10/P50/P90, on-time probability, pessimistic stress test, map/curves and calibration and decision metrics.
The images shown are demo outputs included in the WF20 repository; they are not live navigation, dispatch or safety instructions.
4 · How it works
Traffic-aware alternatives + calibrated uncertainty + risk-aware departure optimisation
1Google Routes API v2 traffic-aware/static durations, alternatives, distance and geometry; hourly Open-Meteo; local time; role/scenario; historical near-departure ETA proxy
2API baseline → bias adjustment with 1–49 samples → Random Forest from 50 samples; P10/P50/P90, on-time probability and latest-departure selection subject to threshold and pessimistic gate
3MAE, RMSE, MAPE, median/p90 absolute error, P10–P90 coverage, top-1 departure accuracy, departure regret and explicit warm-up stage
4Persistence, rendering and governed publishing.
5 · Data and signals
| Element | Detail |
|---|---|
| 5 · Data and signals | Google Routes API v2 traffic-aware/static durations, alternatives, distance and geometry; hourly Open-Meteo; local time; role/scenario; historical near-departure ETA proxy |
| Industrial transferability | commute and logistics planning, route/departure selection and non-operational decision-support demonstrations |
| Research | interval calibration, decision regret, distribution shift and provider stress scenarios |
| Interpretation condition | educational use only: not navigation, dispatch or safety advice; the near-departure ETA proxy is not GPS ground truth |
API baseline → bias adjustment with 1–49 samples → Random Forest from 50 samples; P10/P50/P90, on-time probability and latest-departure selection subject to threshold and pessimistic gate
MAE, RMSE, MAPE, median/p90 absolute error, P10–P90 coverage, top-1 departure accuracy, departure regret and explicit warm-up stage
Python, Google Routes API v2, Open-Meteo, scikit-learn RandomForest, pandas/NumPy, Pillow/Matplotlib, GitHub Actions, AT Protocol/Bluesky
API traffic and the ETA proxy are not GPS ground truth; closures, incidents, simplified shifts and city-scale weather may be missed
Workflow technical dossier
Methods, checks, stack and transferability conditions.
5 · Data and signals
Google Routes API v2 traffic-aware/static durations, alternatives, distance and geometry; hourly Open-Meteo; local time; role/scenario; historical near-departure ETA proxy
Inference / calculation
API baseline → bias adjustment with 1–49 samples → Random Forest from 50 samples; P10/P50/P90, on-time probability and latest-departure selection subject to threshold and pessimistic gate
Validation
MAE, RMSE, MAPE, median/p90 absolute error, P10–P90 coverage, top-1 departure accuracy, departure regret and explicit warm-up stage
Stack
Python, Google Routes API v2, Open-Meteo, scikit-learn RandomForest, pandas/NumPy, Pillow/Matplotlib, GitHub Actions, AT Protocol/Bluesky
Limitations
API traffic and the ETA proxy are not GPS ground truth; closures, incidents, simplified shifts and city-scale weather may be missed
See recent outputs
The Publications page collects local previews and, on request, the public feed.
Potential transferability
Have a problem similar to The Public Service Run?
A first discussion can start from the objective, available data and success criterion without sending confidential information. A result observed here is not assumed to transfer automatically to another context.