A novel Bayesian hierarchical model for road safety hotspot prediction

详细信息    查看全文

• 作者：Lee Fawcett^a ; ^{lee.fawcett@ncl.ac.uk} ; Neil Thorpe^b ; ^{neil.thorpe@ncl.ac.uk} ; Joseph Matthews^a ; ^{j.matthews2@ncl.ac.uk} ; Karsten Kremer^c ; ^{Karsten.Kremer@ptvgroup.com}
• 关键词：Accident prediction models ; Bayesian statistics ; Hotspot prediction ; Predictive distribution ; Regression-to-mean ; Trend
• 刊名：Accident Analysis & Prevention
• 出版年：2017
• 出版时间：February 2017
• 年：2017
• 卷：99
• 期：part_PA
• 页码：262-271
• 全文大小：621 K
• 卷排序：99

文摘

A Bayesian hierarchical model for predicting road safety hotspots is proposed. The model extends the classical empirical Bayes method to include data from multiple time-points. The model incorporates adjustments for regression-to-mean and global/local trend. A variance inflation device gives more recent observations more weight in the model. Validation diagnostics suggest the model predicts extremely well, with a high degree of accuracy.