Accounting for heterogeneity in traffic crash prediction: exploring the usage of a dynamic state-space approach

ABSTRACT

A dynamic state-space model is proposed to predict the crash counts. The outcomes of a multivariate regression model that identifies dynamics relationship between the examined factors and the traffic crashes have been incorporated in the proposed sate-space modes as an initial value to describe the state transition process. The KEF and VBAKF algorithms have been developed to estimate the proposed models and the developed models are referred to as SSMKEF and SSM-VBAKF models, respectively. The findings suggest that the proposed state-space model has better prediction accuracy and robustness with the VBAKF algorithm as the estimation method and the prediction accuracy that measured by RMSD can be improved by 23.81% compared to the KEF algorithm. The findings suggest that the proposed SSM-VBAKF and SSM-KEF models can better address the heterogeneity issues and a significant number of zeros in correlated crash data, and provide sufficient fit to the multivariate correlated crash data.

KEYWORDS:

• Highway safety
• dynamic state-space model
• traffic crashes
• SSM-KEF model
• SSM-VBAKF model

Acknowledgements

Special thanks to TDOT for providing the TRIMS data.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This research is supported by funding provided by the Fundamental Research Funds for the Central Universities [grant number 2019RC027].