<![CDATA[Traffic crashes pose a significant threat to public safety, resulting in varying levels of injury severity and vehicle damage for at-fault and non-fault drivers. This study investigates crash severity and its contributing factors by employing Generalized Estimating Equations (GEE) for statistical analysis and Random Forest (RF) for predictive modeling. Using Japanese traffic crash reports from 2019 to 2022, this study applies resampling techniques to address class imbalance. GEE provides interpretable statistical relationships, while RF enhances predictive accuracy by capturing complex variable interactions.The findings indicate that at-fault drivers are more likely to sustain severe injuries, primarily due to stop sign violations, hazardous impact locations, and unsafe road conditions. Meanwhile, non-fault drivers experience greater vehicle damage severity, particularly in rear-end and side-impact collisions, where they have limited control over crash outcomes. RF achieved 71% accuracy in injury severity prediction and 65% in vehicle damage classification, outperforming traditional statistical models. However, GEE provided interpretable coefficients, confirming the influence of traffic compliance, impact location, and driver demographics on crash severity.These results emphasize the need for stricter enforcement of traffic control compliance, intersection redesigns to reduce side-impact crashes, and age-specific safety interventions. The integration of machine learning in crash risk analysis offers promising advancements for proactive traffic safety strategies. Future research should explore hybrid modeling approaches and incorporate real-time traffic conditions to enhance predictive accuracy and policy applications.]]>
Copyrights © 2025
