Harnessing ambient sensing & naturalistic driving systems to understand links between driving volatility and crash propensity in school zones - A generalized hierarchical mixed logit framework

摘要

With the advent of seemingly unstructured big data, and through seamless integration of computation and physical components, cyber-physical systems (CPS) provide an innovative way to enhance safety and resiliency of transport infrastructure. This study focuses on real-world microscopic driving behavior and its relevance to school zone safety - expanding the capability, usability, and safety of dynamic physical systems through data analytics. Driving behavior and school zone safety is a public health concern. The sequence of instantaneous driving decisions and its variations prior to involvement in safety critical events, defined as driving volatility, can be a leading indicator of safety. By harnessing unique naturalistic data on more than 41,000 normal, crash, and near-crash events featuring over 9.4 million temporal samples of real-world driving, a characterization of volatility in microscopic driving decisions is sought at school and non-school zone locations. A big data analytic methodology is proposed for quantifying driving volatility in microscopic real-world driving decisions. Eight different volatility measures are then linked with detailed event-specific characteristics, health history, driving history/experience, and other factors to examine crash propensity at school zones. A comprehensive yet fully flexible state-of-the-art generalized mixed logit framework is employed to fully account for distinct yet related methodological issues of scale and random heterogeneity, containing multinomial logit, random parameter logit, scaled logit, hierarchical scaled logit, and hierarchical generalized mixed logit as special cases. The results reveal that both for school and non-school locations, drivers exhibited greater intentional volatility prior to safety-critical events. Volatility in positive and negative vehicular jerk in longitudinal and lateral directions associates with increases the probability of unsafe outcomes (crashes or near-crashes) at school zones. A one-unit increase in intentional volatility measured by positive vehicular jerk in longitudinal direction associates with a 0.0528 increase in the probability of crash outcome. Importantly, the effect of negative vehicular jerk (braking) in longitudinal direction on the likelihood of crash outcome is almost double. Methodologically, Hierarchical Generalized Mixed Logit model resulted in best-fit, simultaneously accounting for scale and random heterogeneity. When accounted for separately, more parsimonious models accounting for scale heterogeneity performed comparably to the less parsimonious counterparts accounting for random heterogeneity. Importantly, even after accounting for random heterogeneity, substantial heterogeneity due to a "pure scale-effect" is still observed, underscoring the importance of scale effects in influencing the overall contours of variations in the modeled relationships. The study demonstrates the value of observational study design and big data analytics for understanding extreme driving behaviors in safe vs. unsafe driving outcomes. Implications for designing personalized school zone behavioral countermeasures are discussed.