Investigation of vehicle-bicycle hit-and-run crashes

Although cycling has been promoted around the world as a sustainable mode of transportation, bicyclists are among the most vulnerable road users, subject to high injury and fatality risk. The vehicle-bicycle hit-and-run crashes degrade the morality and result in delays of medical services provided to victims. This paper aims to determine the significant factors that contribute to drivers’ hit-and-run behavior in vehicle-bicycle crashes and their interdependency based on a 6-year crash dataset of Victoria, Australia, with an integrated data mining framework. The framework integrates imbalanced data resampling, near zero variance predictor elimination, learning-based feature extraction with random forest algorithm, and association rule mining. The crash-related features that play the most important role in classifying hit-and-run crashes are identified as collision type, gender, age group, vehicle passengers involved, severity of accident, speed zone, road classification, divided road, region and peak hour. The result of the paper can further provide implications on the policies and counter-measures in order to prevent bicyclists from vehicle-bicycle hit-and-run collisions.

尽管自行车作为可持续交通方式在全球范围内得到推广，但骑行者属于道路通行中最易受伤害的群体之一，面临极高的受伤与致死风险。机动车与自行车碰撞后肇事逃逸事故，不仅败坏社会公序良俗，还会延误对受害者的医疗救治。本研究基于澳大利亚维多利亚州6年的交通事故数据集，结合集成数据挖掘框架，旨在识别机动车与自行车碰撞事故中导致驾驶员肇事逃逸行为的关键影响因素及其相互依存关系。该集成框架涵盖了不平衡数据重采样、近零方差预测变量剔除、基于随机森林（Random Forest）算法的学习型特征提取以及关联规则挖掘等技术环节。经识别，对肇事逃逸事故分类起核心作用的事故相关特征包括碰撞类型、性别、年龄组别、涉事车辆乘客数、事故严重程度、限速区域、道路等级、分隔式道路、区域类型以及高峰时段。本研究结果可为制定预防机动车与自行车碰撞肇事逃逸事故的相关政策与防控措施提供理论依据与实践启示。