Visitors security analysis has historically relied on police-reported crash statistics, usually thought of the “gold commonplace” as a result of they straight correlate with fatalities, accidents, and property harm. Nonetheless, counting on historic crash knowledge for predictive modeling presents vital challenges, as a result of such knowledge is inherently a “lagging” indicator. Additionally, crashes are statistically uncommon occasions on arterial and native roads, so it could actually take years to build up adequate knowledge to determine a sound security profile for a particular street section. This sparsity paired with inconsistent reporting requirements throughout areas complicates the event of strong danger prediction fashions. Proactive security evaluation requires “main” measures: proxies for crash danger that correlate with security outcomes however happen extra ceaselessly than crashes.
In “From Lagging to Main: Validating Onerous Braking Occasions as Excessive-Density Indicators of Section Crash Threat“, we consider the efficacy of hard-braking occasions (HBEs) as a scalable surrogate for crash danger. An HBE is an occasion the place a automobile’s ahead deceleration exceeds a particular threshold (-3m/s²), which we interpret as an evasive maneuver. HBEs facilitate network-wide evaluation as a result of they’re sourced from related automobile knowledge, not like proximity-based surrogates like time-to-collision that ceaselessly necessitate the usage of fastened sensors. We established a statistically vital constructive correlation between the charges of crashes (of any severity degree) and HBE frequency by combining public crash knowledge from Virginia and California with anonymized, aggregated HBE data from the Android Auto platform.