Identifying Edge Cases in Accident Fatalities for Human-Controlled Vehicles via Angle-Based Outlier Detection

Fatal accidents remain a significant challenge to road safety globally, particularly for human-controlled vehicles (SAE Level 1 and 2). Despite advancements in vehicle safety technologies, understanding and addressing edge cases — scenarios where conventional safety measures might fail — are crucial for enhancing safety standards. This study offers method for identification and a detailed analysis of edge cases in accident fatalities. Using a dataset of 39,221 accident cases from 2022, the Angle-Based Outlier Detection (ABOD) method was employed to identify 4,429 edge cases. The dataset underwent thorough preprocessing and feature engineering to accurately reflect the complexities of real-world accidents, including encoding categorical variables, normalizing numerical features, and applying ABOD technique to pinpoint anomalous data points. The effectiveness of the method was evaluated using statistical measures such as ROC AUC scores and confusion matrix. The study highlights critical factors contributing to fatalities in edge cases, revealing previously underappreciated aspects of vehicle safety. The implications for improving road safety are discussed in detail. This research lays the groundwork for future studies on advanced safety systems and automated vehicle technologies, utilizing outlier detection technique to generate actionable insights for enhancing road safety.