A traffic accident dataset for Chattanooga, Tennessee

Area of Science:

• Transportation Engineering
• Data Science
• Traffic Safety

Background:

• Accurate traffic data is crucial for understanding and mitigating accident impacts.
• Existing datasets often lack comprehensive fusion of real-time traffic, environmental, and accident specifics.
• Machine learning approaches require well-structured, annotated data for effective traffic pattern analysis.

Purpose of the Study:

• To present a novel, annotated dataset fusing traffic sensor data, weather conditions, light conditions, and accident details.
• To facilitate machine learning-based analysis, prediction, and detection of traffic patterns during accidents.
• To provide baseline data for normal traffic conditions for comparative analysis.

Main Methods:

• Fusion of data from radar detection sensors, weather, light, and traffic accident records.
• Inclusion of time-series data for traffic speed, flow, and occupancy from accident-proximate sensors (nearest, 5 upstream, 5 downstream).
• Annotation of accident type, date, and time within the dataset.

Main Results:

• A comprehensive dataset covering 6 months (November 2020 - April 2021) of traffic and accident data.
• Detailed records for 361 accidents within the monitored area of Chattanooga, Tennessee.
• Dataset format optimized for machine learning tools, databases, and data workflows.

Conclusions:

• The presented dataset provides a valuable resource for advancing research in traffic accident analysis and prediction.
• This fused dataset supports the development of automated accident detection systems.
• The data enables deeper insights into the relationship between traffic conditions, environmental factors, and accident occurrences.