Modeling crash risk via pre-crash trajectory analysis: Evidence from a matched case-control study

Traffic crashes remain a major public health concern, contributing substantially to global injuries and fatalities. A deeper understanding of pre-crash driving behaviors is therefore essential for improving road safety. This study investigates crash-related driving behavior using a large-scale nationwide naturalistic driving dataset from China comprising 545,052 crash-involved trajectories. A matched case-control design was employed, where each crash trajectory was paired with four non-crash counterparts matched by temporal and spatial factors, and kinematic indicators, including speed, acceleration, jerk, and yaw rate, were extracted from the trajectories. A conditional logistic regression model with nonlinear and interaction terms was then applied to quantify crash risk. Results show that the root mean square of deceleration (dec_rms) is the strongest predictor, with crash odds increasing more than fourfold, while the root mean square of jerk (j_rms) and mean yaw rate (yaw_mean) are also positively associated with crash risk. In contrast, higher mean speed (v_mean) and moderate speed variability (v_cv) are linked to lower crash odds, although the squared term of vcv reveals a U-shaped relationship. Interaction effects further indicate that the influence of acceleration on crash risk varies with driving speed, and that combined braking and lateral maneuvers significantly reshape crash likelihood. These findings highlight the value of trajectory-based kinematic indicators for crash risk assessment and demonstrate their potential for integration into real-time safety monitoring and advanced driver-assistance systems.