An open-source tool to identify active travel from hip-worn accelerometer, GPS and GIS data

Area of Science:

• Public Health
• Epidemiology
• Data Science

Background:

• Active travel offers significant public health benefits, including improved health outcomes and reduced traffic.
• Accurately identifying travel modes in large datasets is a significant challenge for researchers.
• Existing methods struggle to precisely differentiate between various travel modes using sensor data.

Purpose of the Study:

• To develop and validate an open-source tool for quantifying time spent in different travel modes.
• To accurately classify time spent stationary, walking, cycling, using trains, or motorized vehicles.
• To enhance the analysis of physical activity and its relation to travel behavior in large datasets.

Main Methods:

• Utilized accelerometer and GPS data from 326 participants in the Examining Neighbourhood Activities in Built Living Environments in London (ENABLE London) study.
• Developed a supervised machine learning model (gradient boosted tree) trained on 131,537 data points.
• Validated the model using five-fold cross-validation and manual identification of travel modes in independent datasets (ENABLE London and STAMP-2).

Main Results:

• Achieved 97.3% accuracy in identifying travel modes during cross-validation.
• Demonstrated high agreement with manual identification (96.0%) and an independent study (96.5%).
• The tool precisely identifies stationary and walking time, with good precision for train/vehicle time and moderate precision for cycling time.

Conclusions:

• Presents a generalizable, open-source tool for accurately identifying travel modes using accelerometer and GPS data.
• Complements physical activity analyses by differentiating travel modes, aiding research into health behaviors.
• Provides all necessary code for replication and application to other datasets, fostering wider research use.