| Summary: | The heterogenous nature of crash data has always been a challenging barrier in interpreting data processing results and in unrevealing the hidden relationships between crash contributing factors. Traffic researchers aim to eliminate the interruption of insignificant factors , identify influential factors and sort out complicated logic relationships among those factors. Understanding the influential factors leading to traffic accidents is essential for designing effective countermeasures. A method commonly employed to address systematic heterogeneity in crash data is to focus on each subgroup of data. However, this approach neglects the independent relationships among factors, and does not ensure homogeneity within each subgroup. In this project, Latent Class Cluster analysis is applied to segment a whole cyclist crash dataset into homogenous subgroups with meaningful influential factors. The manuscript employs data from recorded crashes involving cyclists from 2008 to 2018 by the police in Toronto, Canada. The analyses demonstrate that dividing cyclists’ crash data into seven clusters most efficiently helps in reducing the systematic heterogeneity of the data and aids in understanding the relationships between socio-demographic characteristics, environmental characteristics, maneuver-related characteristics etc. and further identifying determining reasons for the crashes involving cyclists . Based on the clustering results, some significant factors are studied in detail along with socio-economic backgrounds and some countermeasures are proposed. Overall, this study suggests that a latent class clustering approach is suitable for reducing heterogeneity and revealing important hidden relationships in traffic safety analyses.
|
|---|