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

Smartphones are now equipped with sensors capable of recording vehicle performance data at a very fine temporal resolution in a cost-effective way. In this paper, mobile sensor data from smartphones was used to identify and quantify unsafe driving patterns and their relationship with traffic crash incidences. Statistical models that account for measurement error associated with microscopic traffic measures computed using mobile sensor data were developed. The models with microscopic traffic measures were shown to be statistically better than traditional models that only control for roadway geometry and traffic exposure variables. Also, generalized count models that account for measurement error, spatial dependency effects, and random parameter heterogeneity were found to perform better than standard count models.