Predictive Modeling of Crash Injury Outcomes in AV-Related Accidents

As the advancement of automated vehicles (AVs) prioritizes safety, extensive real-world testing is underway to ensure their secure operation on roads. While there's a prevalent belief that AVs will reduce accident frequency, uncertainties regarding their impact on crash severity persist. The primary objective of this study is to leverage the logistic regression technique to predict the likelihood of injuries occurring in these AV-involved accidents by examining a comprehensive array of independent variables. This study delves into AV crash injuries, analyzing 441 accidents reported from 2014 to 2023. The analysis of the model's performance revealed its strong predictive capabilities, as evidenced by a precision score of 0.93, signifying its ability to make accurate predictions of injuries when it detects a positive outcome. The recall score, at 0.94, underscores the model's capacity to effectively identify true positives and minimize false negatives, thus capturing a significant proportion of actual injuries. Furthermore, the Area Under the Receiver Operating Characteristic Curve exhibited a value of 0.753, indicating the model's reasonable discriminatory power in distinguishing between injury and no-injury outcomes. These metrics demonstrate the model's effectiveness in predicting injury outcomes with high accuracy and reliable discrimination between the two classes. The findings shed light on the intricate dynamics of AV-involved crashes, explicitly focusing on injury outcomes.