Modeling the impact of latent driving patterns on traffic safety using mobile sensor data

Area of Science:

• Transportation engineering
• Data science
• Traffic safety research

Background:

• Smartphones offer cost-effective, high-resolution vehicle performance data collection.
• Existing traffic safety models often lack precision in identifying unsafe driving behaviors.

Purpose of the Study:

• To utilize smartphone mobile sensor data for identifying and quantifying unsafe driving patterns.
• To establish the relationship between these driving patterns and traffic crash incidences.
• To develop advanced statistical models for improved crash prediction.

Main Methods:

• Collected microscopic traffic measures using smartphone mobile sensor data.
• Developed statistical models accounting for measurement error in mobile sensor data.
• Employed generalized count models addressing measurement error, spatial dependency, and parameter heterogeneity.

Main Results:

• Models incorporating microscopic traffic measures from mobile sensors significantly outperformed traditional models.
• Advanced count models demonstrated superior performance over standard models in crash incidence prediction.
• Mobile sensor data provides a more accurate basis for analyzing driving behavior and crash risk.

Conclusions:

• Smartphone sensor data is a viable and effective tool for traffic safety analysis.
• Advanced statistical modeling techniques enhance the prediction of traffic crash incidences.
• This approach offers a more granular understanding of the link between driving behavior and road safety.