AI analysis reveals 3x higher severe crash risk for unbelted drivers
Machine learning model pinpoints restraint non-use as top predictor in tree collisions.
A new machine learning study, 'From Canopy to Collision,' leverages CatBoost, SHAP, and binary logistic regression on the Crash Report Sampling System (CRSS) data from 2020–2023 to quantify risk factors in tree-involved run-off-road crashes. The framework identifies restraint non-use as the most influential predictor—unrestrained occupants are nearly three times more likely to suffer severe outcomes (KA: fatal/incapacitating) due to ejection risk. Vehicle age (reduced crashworthiness), speeding violations (higher impact forces), and driver impairment (reduced control) also show substantial effects. Critical interactions emerge between lighting conditions and vehicle age, speeding and visibility, and road surface and speeding, revealing additive risk effects.
These findings inform safe system interventions: enhanced seat belt enforcement, speed management in low-visibility conditions, and vehicle fleet modernization. The study demonstrates how explainable AI (SHAP) can translate complex crash data into targeted policy recommendations, potentially reducing severe injuries in tree-related collisions. With 30 pages and 10 figures, the research provides a rigorous framework for transportation safety agencies to prioritize resources and design data-driven countermeasures.
- Restraint non-use is the top predictor, tripling the odds of severe injury in tree crashes.
- Vehicle age, speeding, and driver impairment significantly increase crash severity via reduced crashworthiness, impact forces, and control.
- Key interactions: speeding + poor lighting, restraint non-use + older vehicles, and wet roads + speeding compound risks.
Why It Matters
Data-driven safety policies from explainable AI can reduce fatalities in tree-involved run-off-road crashes.