Uncovering contributing factors of police injury in traffic crashes: an association rule mining-guided machine learning approach

Police officers face unique crash risks due to the dynamic and high-pressure nature of their duties. Despite their critical roles in traffic safety and law enforcement, research on police-involved crashes remains limited. This study integrates Association Rule Mining (ARM) with interpretable machine learning to identify and quantify factors associated with police injury outcomes. Ten years of Pennsylvania police-involved crash data (2014–2023) were analysed. ARM identified 1,211 rules and highlighted road type, driving speed, collision manner and pre-crash behaviour as influential factors. Top-ranked rules then guided variable selection for the predictive model, strengthening interpretability and empirical grounding. The results showed higher predicted injury probabilities for crashes on local roads, at higher speeds, and involving head-on or angle collisions. Impaired driving and signal or sign violations were also associated with elevated risk. This transparent, generalisable framework provides actionable evidence for targeted interventions to improve officer safety.