A large-scale graph-augmented traffic dataset for data-driven spatio-temporal traffic analysis.

Area of Science:

• Urban mobility
• Transportation engineering
• Data science

Background:

• Existing traffic datasets are often geographically limited, primarily from California.
• There is a need for large-scale, high-resolution traffic data for advanced urban mobility research.
• Machine learning applications in transportation require diverse and comprehensive datasets.

Purpose of the Study:

• Introduce a novel, large-scale, high-resolution traffic dataset from Madrid, Spain.
• Provide a valuable resource for machine learning tasks in urban mobility and smart city development.
• Enhance the generalization of transportation models by offering geographic diversity.

Main Methods:

• Collected over 1.5 billion traffic records from 5,000+ sensors in Madrid (2015-2024).
• Aggregated key metrics (intensity, occupancy, speed) at 15-minute intervals.
• Developed two graph-based spatial representations of the sensor network.

Main Results:

• A comprehensive dataset covering a decade of urban traffic dynamics.
• Inclusion of spatio-temporal and graph-based data structures.
• Dataset supports machine learning (forecasting, representation learning) and urban planning.

Conclusions:

• The Madrid traffic dataset offers significant geographic diversity, improving model generalization.
• This resource facilitates advanced urban mobility analysis and smart city development.
• Traffic data can serve as a covariate for related domains like air quality monitoring.