Modeling the evolution of pre-crash driving behavior and identifying truck crash risk: A complex network approach

Trucks are essential to road transportation, yet their high crash frequency and severity highlight the urgent need for systematic risk analysis. Existing studies mainly rely on retrospective crash data and short-term risk detection, overlooking the multiscale dynamics underlying crash evolution. To address this gap, this study integrates large-scale truck trajectory data with accident records to construct a multi-timescale analytical framework for understanding the evolution of truck crash risk. Specifically, unsupervised machine learning methods-PCA-K-Means, Latent Dirichlet Allocation (LDA), and DTW-K-Means-were applied to extract long-term driving styles, mid-term speed volatility modes, and short-term speed patterns. A multilayer complex network was then developed to model the dynamic interactions from multi-timescale behavior (long-term, mid-term, and short-term) to crash occurrence, with key network metrics revealing essential propagation mechanisms. Four long-term driving styles, three mid-term volatility modes, and six short-term speed patterns were identified. Among these, the aggressive long-term style, moderate-speed high-volatility mode, and continuous deceleration pattern exhibited the highest Overall Transmission Efficiency (OTE). Notably, the OTE of the moderate-speed high-volatility mode surpassed several short-term patterns, indicating that this mid-term behavior can more rapidly propagate potential risks to crash nodes. By tracing key evolution paths, the study further identified three characteristic evolution patterns and four high-risk pathways associated with specific crash types. Furthermore, aggressive driving behavior and failure to yield appeared in 37.20% and 46.51% of these paths, confirming their dominant contributions to truck crash causation. This research clarifies the multiscale behavioral mechanisms driving truck crashes and offers valuable theoretical and practical insights for proactive traffic safety management.