Graph Neural Networks for Road Safety Modeling: Datasets and Evaluations for Accident Analysis

We consider the problem of traffic accident analysis on a road network based on road network connections and traffic volume. Previous works have designed various deep-learning methods using historical records to predict traffic accident occurrences. However, there is a lack of consensus on how accurate existing methods are, and a fundamental issue is the lack of public accident datasets for comprehensive evaluations. This paper constructs a large-scale, unified dataset of traffic accident records from official reports of various states in …

analysis consensus cs.lg cs.si datasets graph graph neural networks modeling network networks neural networks records road safety safety traffic