Causal V: A five-phase causal mining methodology for real-world accident data

Abstract

Amid rapid advancements in autonomous vehicle technology, traffic accidents due to Autonomous Driving Systems (ADS) flaws persist. The critical task of attributing causality to accidents and forecasting potential incidents based on situational data stands at the forefront of enhancing vehicular safety. Addressing these challenges, this study introduces Casual V, a systematic five-phase V-shape approach to constructing a causal Bayesian network (CBN) from massive real-work crash data. To evaluate the proposed approach, a CBN was constructed from more than 300,000 Maryland accident records. Experiments demonstrate that our method is capable of effectively constructing a CBN that provides deep insights into the causal relationships within vehicular accident data.