Exploratory analysis and prediction of automated vehicles crashes in the United States

Abstract

This study aims to understand and predict the underlying correlations of crashes caused by automated vehicles (AVs) and Vehicles with ADAS (Advanced Driver Assistance Systems). The researchers used a dataset of crashes of vehicle with ADS (Autonomous Driving Systems) and ADAS that occurred in the United States between 2019 and 2022. We analysed the relationship between the crash frequency, which represents the number of reported AV crashes associated with each combination of risk factor levels in the dataset, and various factors such as the type of vehicle involved, the make and model of the vehicle, the year of the vehicle, the state and city where the crash occurred, and the type of roadway. We used machine learning algorithms to predict the relationship between various factors and the frequency of crashes. Results showed that the make of the automated vehicles is strongly correlated with the automated crashes frequency, followed by the injury severity, the level of equipment system (i.e., ADS or ADAS), and the object type in contact with the automated vehicle at the time of the crash, especially AV interacting with VRUs. The support vector machine (SVM) algorithm had the best performance in predicting crash frequency for AVs, with MSE of about 0.38 and MAE of about 0.44. For each of the risk factors, we proposed safety countermeasures that can be adopted to reduce AV crashes, however, a holistic approach that considers a variety of factors is necessary to effectively reduce the risk of AV crashes and improve overall safety on the roads.