Crash Injury Analysis of Autonomous Vehicles: A Data-Driven Study

The proliferation of autonomous vehicles (AV) is inevitable, driven by the rapid evolution of sensing and navigation technologies, diminishing the prevalence of human-operated vehicles. While there's optimism that AVs will lower accident rates, their impact on crash severity remains uncertain. This study aims to build a robust crash injury prediction model to foresee crash injury outcomes and gain insights into the factors contributing to crash injuries. This study examines a dataset comprising AV collision data from 2014 to July 2023. This data was obtained through the California Department of Motor Vehicles, which mandates AV manufacturers to provide reports on disengagements and accidents. The analysis encompasses various independent variables, including road type, intersection characteristics, signal presence, parking provisions, the number of vehicles involved, vehicle modes, vehicle damage, collision types, weather conditions, lighting conditions and road surface quality. The study employs random forest and gradient-boosting predictive models and evaluates their performance using accuracy, precision, recall, and F1-score metrics. The results demonstrate that the random forest model surpassed the gradient boosting model in accuracy, showcasing its superior performance in making precise predictions. The study's findings will contribute to a better understanding of factors influencing crash injury and have the potential to foster the development of enhanced, robust protective measures in AVs.