Cost-sensitive semi-supervised deep learning to assess driving risk by application of naturalistic vehicle trajectories
作者：
Highlights：
• Clustering and class equilibrium are not satisfied in driving risk assessment.
• Multi-stage optimized semi-supervised learning can suppress clustering problems.
• Adaptive over balanced cross entropy can maintain training in an over-balance state.
• The dense clusters of high-risk trajectories were often caused by chain reactions.
• Risk assessment could be realized using only 2.5% of total data with minor errors.
摘要
•Clustering and class equilibrium are not satisfied in driving risk assessment.•Multi-stage optimized semi-supervised learning can suppress clustering problems.•Adaptive over balanced cross entropy can maintain training in an over-balance state.•The dense clusters of high-risk trajectories were often caused by chain reactions.•Risk assessment could be realized using only 2.5% of total data with minor errors.
论文关键词：Driving risk,Semi-supervised deep learning,Cost-sensitive,Naturalistic vehicle trajectories
论文评审过程：Received 16 December 2020, Revised 19 February 2021, Accepted 10 April 2021, Available online 20 April 2021, Version of Record 29 April 2021.
论文官网地址：https://doi.org/10.1016/j.eswa.2021.115041
