Leveraging autonomous vehicles crash narratives to understand the patterns of parking-related crashes

Abstract

Autonomous vehicles (AVs) parking has been a subject of interest from various researchers; however, the focus has been on the parking demand, algorithm, and policies, while the safety aspect has received less attention, perhaps due to the lack of AV crash data. This study evaluated the magnitude and pattern of AV parking-related crashes that occurred between January 2017 and August 2022 in California. The study applied descriptive analysis, unsupervised text mining, and supervised text mining (Support Vector Machine, Naïve Bayes, Logitboost, Random Forest, and Neural network) with resampling techniques. It was indicated that parking-related crashes constitute about 16% of all AV crashes, most of them are likely to impact the AV on the rear or left side. The unsupervised text mining results showed that AVs in the conventional mode of operation, reversing, and parallel parking are among the key themes associated with parking-related crashes. The Support Vector Machine, Logitboost, Random Forest, and Neural network showed relatively high prediction accuracy. The important features from these supervised text mining approaches were conventional mode, reservsing, passenger vehicle, parallel parking, which confirm the preliminary findings in the unsupervised text mining. The implications of the findings to operators and policymakers are included in the study. Findings from this paper could be used to introduce measures to reduce AV parking-related crashes.

Keywords

• autonomous vehicles
• crash narratives
• machine learning
• parking-related crashes
• text network