Autonomous vehicle safety: An advanced bagging classifier technique for crash injury prediction

Abstract

The increasing utilization of autonomous vehicles (AVs) makes it critical to understand and mitigate their involvement in traffic accidents. This study, therefore, addresses a significant gap in the research on AV safety by focusing on predicting the possibility of injuries in AV-involved crashes. The California Department of Motor Vehicles’ comprehensive dataset of accidents that occurred from 2014 to May 2024 was utilized, and advanced machine learning techniques were applied to develop a model capable of predicting the outcomes of accidents involving AVs. The study found that the bagging classifier model outperforms other models in reliably predicting and identifying severe crashes and minimizing misclassification. Evaluations made through precision-recall, validation, and learning curves confirm the model's robustness, ability to generalize across data subsets, and effectiveness in increasing training data. Key predictors of crash severity include the extent of damage to the AV, vehicle type, manufacturer, and presence of a traffic signal. The study highlights the importance of tailored safety measures, robust safety mechanisms, and advanced traffic management systems to mitigate crash severity. The real-world application of this advanced model promises substantial benefits for vehicle manufacturers, urban planners, policymakers, and end-users, and will contribute to safer roadways.