Awan, F. M. (2025) Travel mode detection Glasgow City 2019-2023. [Data Collection]

Not all data is available to download from this page. Usually this is because the dataset is too large or it is restricted in some way.

Collection description

Overview
The Travel Mode Detection (TMD) data product provides a travel mode matrix that quantifies the number of trips made by each transport mode between every origin–destination (OD) pair. Leveraging mobile phone GPS traces alongside annotated travel diary records, TMD enables detailed insights into modal splits at the urban scale for the period 2019–2023.
Data Sources
1. Mobile GPS Data
Anonymised, timestamped location points from the HUQ mobile app, capturing trajectories of consenting users across the UK.

2. Travel Diary Data
Secondary labelled survey data from TravelAI, detailing true travel modes and trip attributes used to train and validate the classification model.

Methodology
1. Pre-processing of GPS Trajectories:
- Filter out low-accuracy GPS points.
- Segment continuous traces into candidate trips using stay‐point detection (≥ 5 min within 500 m radius).

2.Feature Extraction:
- Mobility Features: trip distance, average speed, acceleration patterns, heading changes.
- Spatial Context: land‐use classification around trajectory, proximity to transport infrastructure (bus stops, train stations, green spaces).

3. Model Training:

- Split TravelAI diary data into 80% train / 20% test sets.
- Train a supervised machine-learning classifier (e.g. XGBoost) to predict modes (car, bus, bicycle, train, walk).
- Hyperparameter tuning via cross-validation to optimise precision and recall.

4. Performance on Test Set:
- Accuracy: 95.6%
- Precision: 86.2%
- Recall: 87.2%

5.Application to HUQ Trips
- Apply trained model to all HUQ-detected trips with complete trajectory data.
- Aggregate counts of trips by predicted mode for each OD pair, producing the TMD matrix.

Manual Validation
Because HUQ lacks ground-truth labels, a stratified random sample of detected trips was manually checked for plausibility. “Feasibility”, whether the predicted mode matches realistic expectations given trip characteristics, was used as the metric:
| Mode | Sample Feasibility |
| ------- | ------------------ |
| Bicycle | 52 % |
| Bus | 64 % |
| Car | 92 % |
| Train | 14 % |
| Walk | 4 % |

Lower feasibility for train and walk largely reflects sparse trajectory points and the small sample sizes for these modes.

Access and Restrictions
The TMD product is available for non-commercial academic research under UBDC’s End User Licence. Data requests are subject to safeguarding protocols to protect user privacy.

More Information
- User Guide for Travel Mode Detection Matrix: Comprehensive documentation on data schema, CSV formats, and example queries.

Repository Staff Only: Update this record