Exploring the heterogeneities in vehicle-involved traffic violations at intersections using latent class clustering and partial proportional odds models

Abstract

Vehicle-involved road crashes, mainly caused by traffic violations, result in major property damage and life losses annually. Thus, it is of vital importance to clarify the heterogeneities of various factors in traffic violation severities. This study quantifies the heterogeneities across different clusters. It selects 15 potential factors from six perspectives based on 40,161 police-reported traffic violations at intersections. An unsupervised learning algorithm combining latent class clustering analysis and partial proportional odds model, as well as marginal effects, is applied to clarify the differences. Four clusters are identified as the optimal based on the Akaike information criterion, Bayesian information criterion (BIC), sample-sized adjusted BIC, and entropy-based approach. The results evidenced that the top five influential factors in traffic violations at intersections were legs of entrance, driver age, road type, area type, and signal control. Their maximum absolute values of their marginal effects were more than 35%. Furthermore, there were great heterogeneities across the clusters, especially between cluster 2 and cluster 3. Findings could provide some insightful information to prioritize effective countermeasures to mitigate the losses of traffic violations and improve road safety.

Keywords:

• traffic violation severity
• heterogeneity
• contributing factors
• latent class clustering
• partial proportional odds model
• unsupervised learning algorithm

Acknowledgments

The authors acknowledge the financial support for this study provided by the National Key Research and Development Program of China (Grant No. 2019YFB1600200) and Postgraduate Research & Practice Innovation Program of Jiangsu Province (NO. KYCX21_0129) and are also grateful to Shangyu Public Security Bureau for the data support.